fs/nilfs2/segment.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * segment.c - NILFS segment constructor.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  *
   9  */
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/buffer_head.h>
  13 #include <linux/writeback.h>
  14 #include <linux/bitops.h>
  15 #include <linux/bio.h>
  16 #include <linux/completion.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/backing-dev.h>
  19 #include <linux/freezer.h>
  20 #include <linux/kthread.h>
  21 #include <linux/crc32.h>
  22 #include <linux/pagevec.h>
  23 #include <linux/slab.h>
  24 #include <linux/sched/signal.h>
  25
  26 #include "nilfs.h"
  27 #include "btnode.h"
  28 #include "page.h"
  29 #include "segment.h"
  30 #include "sufile.h"
  31 #include "cpfile.h"
  32 #include "ifile.h"
  33 #include "segbuf.h"
  34
  35
  36 /*
  37  * Segment constructor
  38  */
  39 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  40
  41 #define SC_MAX_SEGDELTA 64   /*
  42                               * Upper limit of the number of segments
  43                               * appended in collection retry loop
  44                               */
  45
  46 /* Construction mode */
  47 enum {
  48         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  49         SC_LSEG_DSYNC,  /*
  50                          * Flush data blocks of a given file and make
  51                          * a logical segment without a super root.
  52                          */
  53         SC_FLUSH_FILE,  /*
  54                          * Flush data files, leads to segment writes without
  55                          * creating a checkpoint.
  56                          */
  57         SC_FLUSH_DAT,   /*
  58                          * Flush DAT file.  This also creates segments
  59                          * without a checkpoint.
  60                          */
  61 };
  62
  63 /* Stage numbers of dirty block collection */
  64 enum {
  65         NILFS_ST_INIT = 0,
  66         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  67         NILFS_ST_FILE,
  68         NILFS_ST_IFILE,
  69         NILFS_ST_CPFILE,
  70         NILFS_ST_SUFILE,
  71         NILFS_ST_DAT,
  72         NILFS_ST_SR,            /* Super root */
  73         NILFS_ST_DSYNC,         /* Data sync blocks */
  74         NILFS_ST_DONE,
  75 };
  76
  77 #define CREATE_TRACE_POINTS
  78 #include <trace/events/nilfs2.h>
  79
  80 /*
  81  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
  82  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
  83  * the variable must use them because transition of stage count must involve
  84  * trace events (trace_nilfs2_collection_stage_transition).
  85  *
  86  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
  87  * produce tracepoint events. It is provided just for making the intention
  88  * clear.
  89  */
  90 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
  91 {
  92         sci->sc_stage.scnt++;
  93         trace_nilfs2_collection_stage_transition(sci);
  94 }
  95
  96 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
  97 {
  98         sci->sc_stage.scnt = next_scnt;
  99         trace_nilfs2_collection_stage_transition(sci);
 100 }
 101
 102 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
 103 {
 104         return sci->sc_stage.scnt;
 105 }
 106
 107 /* State flags of collection */
 108 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
 109 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
 110 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
 111 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
 112
 113 /* Operations depending on the construction mode and file type */
 114 struct nilfs_sc_operations {
 115         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
 116                             struct inode *);
 117         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
 118                             struct inode *);
 119         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
 120                             struct inode *);
 121         void (*write_data_binfo)(struct nilfs_sc_info *,
 122                                  struct nilfs_segsum_pointer *,
 123                                  union nilfs_binfo *);
 124         void (*write_node_binfo)(struct nilfs_sc_info *,
 125                                  struct nilfs_segsum_pointer *,
 126                                  union nilfs_binfo *);
 127 };
 128
 129 /*
 130  * Other definitions
 131  */
 132 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 133 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 134 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 135 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
 136
 137 #define nilfs_cnt32_gt(a, b)   \
 138         (typecheck(__u32, a) && typecheck(__u32, b) && \
 139          ((__s32)(b) - (__s32)(a) < 0))
 140 #define nilfs_cnt32_ge(a, b)   \
 141         (typecheck(__u32, a) && typecheck(__u32, b) && \
 142          ((__s32)(a) - (__s32)(b) >= 0))
 143 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 144 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 145
 146 static int nilfs_prepare_segment_lock(struct super_block *sb,
 147                                       struct nilfs_transaction_info *ti)
 148 {
 149         struct nilfs_transaction_info *cur_ti = current->journal_info;
 150         void *save = NULL;
 151
 152         if (cur_ti) {
 153                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 154                         return ++cur_ti->ti_count;
 155
 156                 /*
 157                  * If journal_info field is occupied by other FS,
 158                  * it is saved and will be restored on
 159                  * nilfs_transaction_commit().
 160                  */
 161                 nilfs_warn(sb, "journal info from a different FS");
 162                 save = current->journal_info;
 163         }
 164         if (!ti) {
 165                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 166                 if (!ti)
 167                         return -ENOMEM;
 168                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 169         } else {
 170                 ti->ti_flags = 0;
 171         }
 172         ti->ti_count = 0;
 173         ti->ti_save = save;
 174         ti->ti_magic = NILFS_TI_MAGIC;
 175         current->journal_info = ti;
 176         return 0;
 177 }
 178
 179 /**
 180  * nilfs_transaction_begin - start indivisible file operations.
 181  * @sb: super block
 182  * @ti: nilfs_transaction_info
 183  * @vacancy_check: flags for vacancy rate checks
 184  *
 185  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 186  * the segment semaphore, to make a segment construction and write tasks
 187  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 188  * The region enclosed by these two functions can be nested.  To avoid a
 189  * deadlock, the semaphore is only acquired or released in the outermost call.
 190  *
 191  * This function allocates a nilfs_transaction_info struct to keep context
 192  * information on it.  It is initialized and hooked onto the current task in
 193  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 194  * instead; otherwise a new struct is assigned from a slab.
 195  *
 196  * When @vacancy_check flag is set, this function will check the amount of
 197  * free space, and will wait for the GC to reclaim disk space if low capacity.
 198  *
 199  * Return Value: On success, 0 is returned. On error, one of the following
 200  * negative error code is returned.
 201  *
 202  * %-ENOMEM - Insufficient memory available.
 203  *
 204  * %-ENOSPC - No space left on device
 205  */
 206 int nilfs_transaction_begin(struct super_block *sb,
 207                             struct nilfs_transaction_info *ti,
 208                             int vacancy_check)
 209 {
 210         struct the_nilfs *nilfs;
 211         int ret = nilfs_prepare_segment_lock(sb, ti);
 212         struct nilfs_transaction_info *trace_ti;
 213
 214         if (unlikely(ret < 0))
 215                 return ret;
 216         if (ret > 0) {
 217                 trace_ti = current->journal_info;
 218
 219                 trace_nilfs2_transaction_transition(sb, trace_ti,
 220                                     trace_ti->ti_count, trace_ti->ti_flags,
 221                                     TRACE_NILFS2_TRANSACTION_BEGIN);
 222                 return 0;
 223         }
 224
 225         sb_start_intwrite(sb);
 226
 227         nilfs = sb->s_fs_info;
 228         down_read(&nilfs->ns_segctor_sem);
 229         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 230                 up_read(&nilfs->ns_segctor_sem);
 231                 ret = -ENOSPC;
 232                 goto failed;
 233         }
 234
 235         trace_ti = current->journal_info;
 236         trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
 237                                             trace_ti->ti_flags,
 238                                             TRACE_NILFS2_TRANSACTION_BEGIN);
 239         return 0;
 240
 241  failed:
 242         ti = current->journal_info;
 243         current->journal_info = ti->ti_save;
 244         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 245                 kmem_cache_free(nilfs_transaction_cachep, ti);
 246         sb_end_intwrite(sb);
 247         return ret;
 248 }
 249
 250 /**
 251  * nilfs_transaction_commit - commit indivisible file operations.
 252  * @sb: super block
 253  *
 254  * nilfs_transaction_commit() releases the read semaphore which is
 255  * acquired by nilfs_transaction_begin(). This is only performed
 256  * in outermost call of this function.  If a commit flag is set,
 257  * nilfs_transaction_commit() sets a timer to start the segment
 258  * constructor.  If a sync flag is set, it starts construction
 259  * directly.
 260  */
 261 int nilfs_transaction_commit(struct super_block *sb)
 262 {
 263         struct nilfs_transaction_info *ti = current->journal_info;
 264         struct the_nilfs *nilfs = sb->s_fs_info;
 265         int err = 0;
 266
 267         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 268         ti->ti_flags |= NILFS_TI_COMMIT;
 269         if (ti->ti_count > 0) {
 270                 ti->ti_count--;
 271                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 272                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 273                 return 0;
 274         }
 275         if (nilfs->ns_writer) {
 276                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 277
 278                 if (ti->ti_flags & NILFS_TI_COMMIT)
 279                         nilfs_segctor_start_timer(sci);
 280                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 281                         nilfs_segctor_do_flush(sci, 0);
 282         }
 283         up_read(&nilfs->ns_segctor_sem);
 284         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 285                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 286
 287         current->journal_info = ti->ti_save;
 288
 289         if (ti->ti_flags & NILFS_TI_SYNC)
 290                 err = nilfs_construct_segment(sb);
 291         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 292                 kmem_cache_free(nilfs_transaction_cachep, ti);
 293         sb_end_intwrite(sb);
 294         return err;
 295 }
 296
 297 void nilfs_transaction_abort(struct super_block *sb)
 298 {
 299         struct nilfs_transaction_info *ti = current->journal_info;
 300         struct the_nilfs *nilfs = sb->s_fs_info;
 301
 302         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 303         if (ti->ti_count > 0) {
 304                 ti->ti_count--;
 305                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 306                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 307                 return;
 308         }
 309         up_read(&nilfs->ns_segctor_sem);
 310
 311         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 312                     ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 313
 314         current->journal_info = ti->ti_save;
 315         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 316                 kmem_cache_free(nilfs_transaction_cachep, ti);
 317         sb_end_intwrite(sb);
 318 }
 319
 320 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 321 {
 322         struct the_nilfs *nilfs = sb->s_fs_info;
 323         struct nilfs_sc_info *sci = nilfs->ns_writer;
 324
 325         if (!sci || !sci->sc_flush_request)
 326                 return;
 327
 328         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 329         up_read(&nilfs->ns_segctor_sem);
 330
 331         down_write(&nilfs->ns_segctor_sem);
 332         if (sci->sc_flush_request &&
 333             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 334                 struct nilfs_transaction_info *ti = current->journal_info;
 335
 336                 ti->ti_flags |= NILFS_TI_WRITER;
 337                 nilfs_segctor_do_immediate_flush(sci);
 338                 ti->ti_flags &= ~NILFS_TI_WRITER;
 339         }
 340         downgrade_write(&nilfs->ns_segctor_sem);
 341 }
 342
 343 static void nilfs_transaction_lock(struct super_block *sb,
 344                                    struct nilfs_transaction_info *ti,
 345                                    int gcflag)
 346 {
 347         struct nilfs_transaction_info *cur_ti = current->journal_info;
 348         struct the_nilfs *nilfs = sb->s_fs_info;
 349         struct nilfs_sc_info *sci = nilfs->ns_writer;
 350
 351         WARN_ON(cur_ti);
 352         ti->ti_flags = NILFS_TI_WRITER;
 353         ti->ti_count = 0;
 354         ti->ti_save = cur_ti;
 355         ti->ti_magic = NILFS_TI_MAGIC;
 356         current->journal_info = ti;
 357
 358         for (;;) {
 359                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 360                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
 361
 362                 down_write(&nilfs->ns_segctor_sem);
 363                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 364                         break;
 365
 366                 nilfs_segctor_do_immediate_flush(sci);
 367
 368                 up_write(&nilfs->ns_segctor_sem);
 369                 cond_resched();
 370         }
 371         if (gcflag)
 372                 ti->ti_flags |= NILFS_TI_GC;
 373
 374         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 375                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
 376 }
 377
 378 static void nilfs_transaction_unlock(struct super_block *sb)
 379 {
 380         struct nilfs_transaction_info *ti = current->journal_info;
 381         struct the_nilfs *nilfs = sb->s_fs_info;
 382
 383         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 384         BUG_ON(ti->ti_count > 0);
 385
 386         up_write(&nilfs->ns_segctor_sem);
 387         current->journal_info = ti->ti_save;
 388
 389         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 390                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
 391 }
 392
 393 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 394                                             struct nilfs_segsum_pointer *ssp,
 395                                             unsigned int bytes)
 396 {
 397         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 398         unsigned int blocksize = sci->sc_super->s_blocksize;
 399         void *p;
 400
 401         if (unlikely(ssp->offset + bytes > blocksize)) {
 402                 ssp->offset = 0;
 403                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 404                                                &segbuf->sb_segsum_buffers));
 405                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 406         }
 407         p = ssp->bh->b_data + ssp->offset;
 408         ssp->offset += bytes;
 409         return p;
 410 }
 411
 412 /**
 413  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 414  * @sci: nilfs_sc_info
 415  */
 416 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 417 {
 418         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 419         struct buffer_head *sumbh;
 420         unsigned int sumbytes;
 421         unsigned int flags = 0;
 422         int err;
 423
 424         if (nilfs_doing_gc())
 425                 flags = NILFS_SS_GC;
 426         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 427         if (unlikely(err))
 428                 return err;
 429
 430         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 431         sumbytes = segbuf->sb_sum.sumbytes;
 432         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 433         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 434         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 435         return 0;
 436 }
 437
 438 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 439 {
 440         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 441         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 442                 return -E2BIG; /*
 443                                 * The current segment is filled up
 444                                 * (internal code)
 445                                 */
 446         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 447         return nilfs_segctor_reset_segment_buffer(sci);
 448 }
 449
 450 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 451 {
 452         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 453         int err;
 454
 455         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 456                 err = nilfs_segctor_feed_segment(sci);
 457                 if (err)
 458                         return err;
 459                 segbuf = sci->sc_curseg;
 460         }
 461         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 462         if (likely(!err))
 463                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 464         return err;
 465 }
 466
 467 /*
 468  * Functions for making segment summary and payloads
 469  */
 470 static int nilfs_segctor_segsum_block_required(
 471         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 472         unsigned int binfo_size)
 473 {
 474         unsigned int blocksize = sci->sc_super->s_blocksize;
 475         /* Size of finfo and binfo is enough small against blocksize */
 476
 477         return ssp->offset + binfo_size +
 478                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 479                 blocksize;
 480 }
 481
 482 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 483                                       struct inode *inode)
 484 {
 485         sci->sc_curseg->sb_sum.nfinfo++;
 486         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 487         nilfs_segctor_map_segsum_entry(
 488                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 489
 490         if (NILFS_I(inode)->i_root &&
 491             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 492                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 493         /* skip finfo */
 494 }
 495
 496 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 497                                     struct inode *inode)
 498 {
 499         struct nilfs_finfo *finfo;
 500         struct nilfs_inode_info *ii;
 501         struct nilfs_segment_buffer *segbuf;
 502         __u64 cno;
 503
 504         if (sci->sc_blk_cnt == 0)
 505                 return;
 506
 507         ii = NILFS_I(inode);
 508
 509         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 510                 cno = ii->i_cno;
 511         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 512                 cno = 0;
 513         else
 514                 cno = sci->sc_cno;
 515
 516         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 517                                                  sizeof(*finfo));
 518         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 519         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 520         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 521         finfo->fi_cno = cpu_to_le64(cno);
 522
 523         segbuf = sci->sc_curseg;
 524         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 525                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 526         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 527         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 528 }
 529
 530 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 531                                         struct buffer_head *bh,
 532                                         struct inode *inode,
 533                                         unsigned int binfo_size)
 534 {
 535         struct nilfs_segment_buffer *segbuf;
 536         int required, err = 0;
 537
 538  retry:
 539         segbuf = sci->sc_curseg;
 540         required = nilfs_segctor_segsum_block_required(
 541                 sci, &sci->sc_binfo_ptr, binfo_size);
 542         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 543                 nilfs_segctor_end_finfo(sci, inode);
 544                 err = nilfs_segctor_feed_segment(sci);
 545                 if (err)
 546                         return err;
 547                 goto retry;
 548         }
 549         if (unlikely(required)) {
 550                 err = nilfs_segbuf_extend_segsum(segbuf);
 551                 if (unlikely(err))
 552                         goto failed;
 553         }
 554         if (sci->sc_blk_cnt == 0)
 555                 nilfs_segctor_begin_finfo(sci, inode);
 556
 557         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 558         /* Substitution to vblocknr is delayed until update_blocknr() */
 559         nilfs_segbuf_add_file_buffer(segbuf, bh);
 560         sci->sc_blk_cnt++;
 561  failed:
 562         return err;
 563 }
 564
 565 /*
 566  * Callback functions that enumerate, mark, and collect dirty blocks
 567  */
 568 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 569                                    struct buffer_head *bh, struct inode *inode)
 570 {
 571         int err;
 572
 573         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 574         if (err < 0)
 575                 return err;
 576
 577         err = nilfs_segctor_add_file_block(sci, bh, inode,
 578                                            sizeof(struct nilfs_binfo_v));
 579         if (!err)
 580                 sci->sc_datablk_cnt++;
 581         return err;
 582 }
 583
 584 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 585                                    struct buffer_head *bh,
 586                                    struct inode *inode)
 587 {
 588         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 589 }
 590
 591 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 592                                    struct buffer_head *bh,
 593                                    struct inode *inode)
 594 {
 595         WARN_ON(!buffer_dirty(bh));
 596         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 597 }
 598
 599 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 600                                         struct nilfs_segsum_pointer *ssp,
 601                                         union nilfs_binfo *binfo)
 602 {
 603         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 604                 sci, ssp, sizeof(*binfo_v));
 605         *binfo_v = binfo->bi_v;
 606 }
 607
 608 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 609                                         struct nilfs_segsum_pointer *ssp,
 610                                         union nilfs_binfo *binfo)
 611 {
 612         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 613                 sci, ssp, sizeof(*vblocknr));
 614         *vblocknr = binfo->bi_v.bi_vblocknr;
 615 }
 616
 617 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
 618         .collect_data = nilfs_collect_file_data,
 619         .collect_node = nilfs_collect_file_node,
 620         .collect_bmap = nilfs_collect_file_bmap,
 621         .write_data_binfo = nilfs_write_file_data_binfo,
 622         .write_node_binfo = nilfs_write_file_node_binfo,
 623 };
 624
 625 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 626                                   struct buffer_head *bh, struct inode *inode)
 627 {
 628         int err;
 629
 630         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 631         if (err < 0)
 632                 return err;
 633
 634         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 635         if (!err)
 636                 sci->sc_datablk_cnt++;
 637         return err;
 638 }
 639
 640 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 641                                   struct buffer_head *bh, struct inode *inode)
 642 {
 643         WARN_ON(!buffer_dirty(bh));
 644         return nilfs_segctor_add_file_block(sci, bh, inode,
 645                                             sizeof(struct nilfs_binfo_dat));
 646 }
 647
 648 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 649                                        struct nilfs_segsum_pointer *ssp,
 650                                        union nilfs_binfo *binfo)
 651 {
 652         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 653                                                           sizeof(*blkoff));
 654         *blkoff = binfo->bi_dat.bi_blkoff;
 655 }
 656
 657 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 658                                        struct nilfs_segsum_pointer *ssp,
 659                                        union nilfs_binfo *binfo)
 660 {
 661         struct nilfs_binfo_dat *binfo_dat =
 662                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 663         *binfo_dat = binfo->bi_dat;
 664 }
 665
 666 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
 667         .collect_data = nilfs_collect_dat_data,
 668         .collect_node = nilfs_collect_file_node,
 669         .collect_bmap = nilfs_collect_dat_bmap,
 670         .write_data_binfo = nilfs_write_dat_data_binfo,
 671         .write_node_binfo = nilfs_write_dat_node_binfo,
 672 };
 673
 674 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 675         .collect_data = nilfs_collect_file_data,
 676         .collect_node = NULL,
 677         .collect_bmap = NULL,
 678         .write_data_binfo = nilfs_write_file_data_binfo,
 679         .write_node_binfo = NULL,
 680 };
 681
 682 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 683                                               struct list_head *listp,
 684                                               size_t nlimit,
 685                                               loff_t start, loff_t end)
 686 {
 687         struct address_space *mapping = inode->i_mapping;
 688         struct pagevec pvec;
 689         pgoff_t index = 0, last = ULONG_MAX;
 690         size_t ndirties = 0;
 691         int i;
 692
 693         if (unlikely(start != 0 || end != LLONG_MAX)) {
 694                 /*
 695                  * A valid range is given for sync-ing data pages. The
 696                  * range is rounded to per-page; extra dirty buffers
 697                  * may be included if blocksize < pagesize.
 698                  */
 699                 index = start >> PAGE_SHIFT;
 700                 last = end >> PAGE_SHIFT;
 701         }
 702         pagevec_init(&pvec);
 703  repeat:
 704         if (unlikely(index > last) ||
 705             !pagevec_lookup_range_tag(&pvec, mapping, &index, last,
 706                                 PAGECACHE_TAG_DIRTY))
 707                 return ndirties;
 708
 709         for (i = 0; i < pagevec_count(&pvec); i++) {
 710                 struct buffer_head *bh, *head;
 711                 struct page *page = pvec.pages[i];
 712
 713                 lock_page(page);
 714                 if (!page_has_buffers(page))
 715                         create_empty_buffers(page, i_blocksize(inode), 0);
 716                 unlock_page(page);
 717
 718                 bh = head = page_buffers(page);
 719                 do {
 720                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 721                                 continue;
 722                         get_bh(bh);
 723                         list_add_tail(&bh->b_assoc_buffers, listp);
 724                         ndirties++;
 725                         if (unlikely(ndirties >= nlimit)) {
 726                                 pagevec_release(&pvec);
 727                                 cond_resched();
 728                                 return ndirties;
 729                         }
 730                 } while (bh = bh->b_this_page, bh != head);
 731         }
 732         pagevec_release(&pvec);
 733         cond_resched();
 734         goto repeat;
 735 }
 736
 737 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 738                                             struct list_head *listp)
 739 {
 740         struct nilfs_inode_info *ii = NILFS_I(inode);
 741         struct address_space *mapping = &ii->i_btnode_cache;
 742         struct pagevec pvec;
 743         struct buffer_head *bh, *head;
 744         unsigned int i;
 745         pgoff_t index = 0;
 746
 747         pagevec_init(&pvec);
 748
 749         while (pagevec_lookup_tag(&pvec, mapping, &index,
 750                                         PAGECACHE_TAG_DIRTY)) {
 751                 for (i = 0; i < pagevec_count(&pvec); i++) {
 752                         bh = head = page_buffers(pvec.pages[i]);
 753                         do {
 754                                 if (buffer_dirty(bh) &&
 755                                                 !buffer_async_write(bh)) {
 756                                         get_bh(bh);
 757                                         list_add_tail(&bh->b_assoc_buffers,
 758                                                       listp);
 759                                 }
 760                                 bh = bh->b_this_page;
 761                         } while (bh != head);
 762                 }
 763                 pagevec_release(&pvec);
 764                 cond_resched();
 765         }
 766 }
 767
 768 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 769                                struct list_head *head, int force)
 770 {
 771         struct nilfs_inode_info *ii, *n;
 772         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 773         unsigned int nv = 0;
 774
 775         while (!list_empty(head)) {
 776                 spin_lock(&nilfs->ns_inode_lock);
 777                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 778                         list_del_init(&ii->i_dirty);
 779                         if (force) {
 780                                 if (unlikely(ii->i_bh)) {
 781                                         brelse(ii->i_bh);
 782                                         ii->i_bh = NULL;
 783                                 }
 784                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 785                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 786                                 list_add_tail(&ii->i_dirty,
 787                                               &nilfs->ns_dirty_files);
 788                                 continue;
 789                         }
 790                         ivec[nv++] = ii;
 791                         if (nv == SC_N_INODEVEC)
 792                                 break;
 793                 }
 794                 spin_unlock(&nilfs->ns_inode_lock);
 795
 796                 for (pii = ivec; nv > 0; pii++, nv--)
 797                         iput(&(*pii)->vfs_inode);
 798         }
 799 }
 800
 801 static void nilfs_iput_work_func(struct work_struct *work)
 802 {
 803         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 804                                                  sc_iput_work);
 805         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 806
 807         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 808 }
 809
 810 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 811                                      struct nilfs_root *root)
 812 {
 813         int ret = 0;
 814
 815         if (nilfs_mdt_fetch_dirty(root->ifile))
 816                 ret++;
 817         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 818                 ret++;
 819         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 820                 ret++;
 821         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 822                 ret++;
 823         return ret;
 824 }
 825
 826 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 827 {
 828         return list_empty(&sci->sc_dirty_files) &&
 829                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 830                 sci->sc_nfreesegs == 0 &&
 831                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 832 }
 833
 834 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 835 {
 836         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 837         int ret = 0;
 838
 839         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 840                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 841
 842         spin_lock(&nilfs->ns_inode_lock);
 843         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 844                 ret++;
 845
 846         spin_unlock(&nilfs->ns_inode_lock);
 847         return ret;
 848 }
 849
 850 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 851 {
 852         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 853
 854         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 855         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 856         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 857         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 858 }
 859
 860 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 861 {
 862         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 863         struct buffer_head *bh_cp;
 864         struct nilfs_checkpoint *raw_cp;
 865         int err;
 866
 867         /* XXX: this interface will be changed */
 868         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 869                                           &raw_cp, &bh_cp);
 870         if (likely(!err)) {
 871                 /*
 872                  * The following code is duplicated with cpfile.  But, it is
 873                  * needed to collect the checkpoint even if it was not newly
 874                  * created.
 875                  */
 876                 mark_buffer_dirty(bh_cp);
 877                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 878                 nilfs_cpfile_put_checkpoint(
 879                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 880         } else
 881                 WARN_ON(err == -EINVAL || err == -ENOENT);
 882
 883         return err;
 884 }
 885
 886 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 887 {
 888         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 889         struct buffer_head *bh_cp;
 890         struct nilfs_checkpoint *raw_cp;
 891         int err;
 892
 893         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 894                                           &raw_cp, &bh_cp);
 895         if (unlikely(err)) {
 896                 WARN_ON(err == -EINVAL || err == -ENOENT);
 897                 goto failed_ibh;
 898         }
 899         raw_cp->cp_snapshot_list.ssl_next = 0;
 900         raw_cp->cp_snapshot_list.ssl_prev = 0;
 901         raw_cp->cp_inodes_count =
 902                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 903         raw_cp->cp_blocks_count =
 904                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 905         raw_cp->cp_nblk_inc =
 906                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 907         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 908         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 909
 910         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 911                 nilfs_checkpoint_clear_minor(raw_cp);
 912         else
 913                 nilfs_checkpoint_set_minor(raw_cp);
 914
 915         nilfs_write_inode_common(sci->sc_root->ifile,
 916                                  &raw_cp->cp_ifile_inode, 1);
 917         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 918         return 0;
 919
 920  failed_ibh:
 921         return err;
 922 }
 923
 924 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 925                                     struct nilfs_inode_info *ii)
 926
 927 {
 928         struct buffer_head *ibh;
 929         struct nilfs_inode *raw_inode;
 930
 931         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 932                 ibh = ii->i_bh;
 933                 BUG_ON(!ibh);
 934                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 935                                                   ibh);
 936                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 937                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 938         }
 939 }
 940
 941 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 942 {
 943         struct nilfs_inode_info *ii;
 944
 945         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 946                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 947                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 948         }
 949 }
 950
 951 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 952                                              struct the_nilfs *nilfs)
 953 {
 954         struct buffer_head *bh_sr;
 955         struct nilfs_super_root *raw_sr;
 956         unsigned int isz, srsz;
 957
 958         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 959         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 960         isz = nilfs->ns_inode_size;
 961         srsz = NILFS_SR_BYTES(isz);
 962
 963         raw_sr->sr_bytes = cpu_to_le16(srsz);
 964         raw_sr->sr_nongc_ctime
 965                 = cpu_to_le64(nilfs_doing_gc() ?
 966                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 967         raw_sr->sr_flags = 0;
 968
 969         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 970                                  NILFS_SR_DAT_OFFSET(isz), 1);
 971         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 972                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 973         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 974                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 975         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 976 }
 977
 978 static void nilfs_redirty_inodes(struct list_head *head)
 979 {
 980         struct nilfs_inode_info *ii;
 981
 982         list_for_each_entry(ii, head, i_dirty) {
 983                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 984                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 985         }
 986 }
 987
 988 static void nilfs_drop_collected_inodes(struct list_head *head)
 989 {
 990         struct nilfs_inode_info *ii;
 991
 992         list_for_each_entry(ii, head, i_dirty) {
 993                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 994                         continue;
 995
 996                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 997                 set_bit(NILFS_I_UPDATED, &ii->i_state);
 998         }
 999 }
1000
1001 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1002                                        struct inode *inode,
1003                                        struct list_head *listp,
1004                                        int (*collect)(struct nilfs_sc_info *,
1005                                                       struct buffer_head *,
1006                                                       struct inode *))
1007 {
1008         struct buffer_head *bh, *n;
1009         int err = 0;
1010
1011         if (collect) {
1012                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1013                         list_del_init(&bh->b_assoc_buffers);
1014                         err = collect(sci, bh, inode);
1015                         brelse(bh);
1016                         if (unlikely(err))
1017                                 goto dispose_buffers;
1018                 }
1019                 return 0;
1020         }
1021
1022  dispose_buffers:
1023         while (!list_empty(listp)) {
1024                 bh = list_first_entry(listp, struct buffer_head,
1025                                       b_assoc_buffers);
1026                 list_del_init(&bh->b_assoc_buffers);
1027                 brelse(bh);
1028         }
1029         return err;
1030 }
1031
1032 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1033 {
1034         /* Remaining number of blocks within segment buffer */
1035         return sci->sc_segbuf_nblocks -
1036                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1037 }
1038
1039 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1040                                    struct inode *inode,
1041                                    const struct nilfs_sc_operations *sc_ops)
1042 {
1043         LIST_HEAD(data_buffers);
1044         LIST_HEAD(node_buffers);
1045         int err;
1046
1047         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1048                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1049
1050                 n = nilfs_lookup_dirty_data_buffers(
1051                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1052                 if (n > rest) {
1053                         err = nilfs_segctor_apply_buffers(
1054                                 sci, inode, &data_buffers,
1055                                 sc_ops->collect_data);
1056                         BUG_ON(!err); /* always receive -E2BIG or true error */
1057                         goto break_or_fail;
1058                 }
1059         }
1060         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1061
1062         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1063                 err = nilfs_segctor_apply_buffers(
1064                         sci, inode, &data_buffers, sc_ops->collect_data);
1065                 if (unlikely(err)) {
1066                         /* dispose node list */
1067                         nilfs_segctor_apply_buffers(
1068                                 sci, inode, &node_buffers, NULL);
1069                         goto break_or_fail;
1070                 }
1071                 sci->sc_stage.flags |= NILFS_CF_NODE;
1072         }
1073         /* Collect node */
1074         err = nilfs_segctor_apply_buffers(
1075                 sci, inode, &node_buffers, sc_ops->collect_node);
1076         if (unlikely(err))
1077                 goto break_or_fail;
1078
1079         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1080         err = nilfs_segctor_apply_buffers(
1081                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1082         if (unlikely(err))
1083                 goto break_or_fail;
1084
1085         nilfs_segctor_end_finfo(sci, inode);
1086         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1087
1088  break_or_fail:
1089         return err;
1090 }
1091
1092 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1093                                          struct inode *inode)
1094 {
1095         LIST_HEAD(data_buffers);
1096         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1097         int err;
1098
1099         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1100                                             sci->sc_dsync_start,
1101                                             sci->sc_dsync_end);
1102
1103         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1104                                           nilfs_collect_file_data);
1105         if (!err) {
1106                 nilfs_segctor_end_finfo(sci, inode);
1107                 BUG_ON(n > rest);
1108                 /* always receive -E2BIG or true error if n > rest */
1109         }
1110         return err;
1111 }
1112
1113 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1114 {
1115         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1116         struct list_head *head;
1117         struct nilfs_inode_info *ii;
1118         size_t ndone;
1119         int err = 0;
1120
1121         switch (nilfs_sc_cstage_get(sci)) {
1122         case NILFS_ST_INIT:
1123                 /* Pre-processes */
1124                 sci->sc_stage.flags = 0;
1125
1126                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1127                         sci->sc_nblk_inc = 0;
1128                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1129                         if (mode == SC_LSEG_DSYNC) {
1130                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1131                                 goto dsync_mode;
1132                         }
1133                 }
1134
1135                 sci->sc_stage.dirty_file_ptr = NULL;
1136                 sci->sc_stage.gc_inode_ptr = NULL;
1137                 if (mode == SC_FLUSH_DAT) {
1138                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1139                         goto dat_stage;
1140                 }
1141                 nilfs_sc_cstage_inc(sci);
1142                 fallthrough;
1143         case NILFS_ST_GC:
1144                 if (nilfs_doing_gc()) {
1145                         head = &sci->sc_gc_inodes;
1146                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1147                                                 head, i_dirty);
1148                         list_for_each_entry_continue(ii, head, i_dirty) {
1149                                 err = nilfs_segctor_scan_file(
1150                                         sci, &ii->vfs_inode,
1151                                         &nilfs_sc_file_ops);
1152                                 if (unlikely(err)) {
1153                                         sci->sc_stage.gc_inode_ptr = list_entry(
1154                                                 ii->i_dirty.prev,
1155                                                 struct nilfs_inode_info,
1156                                                 i_dirty);
1157                                         goto break_or_fail;
1158                                 }
1159                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1160                         }
1161                         sci->sc_stage.gc_inode_ptr = NULL;
1162                 }
1163                 nilfs_sc_cstage_inc(sci);
1164                 fallthrough;
1165         case NILFS_ST_FILE:
1166                 head = &sci->sc_dirty_files;
1167                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1168                                         i_dirty);
1169                 list_for_each_entry_continue(ii, head, i_dirty) {
1170                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1171
1172                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1173                                                       &nilfs_sc_file_ops);
1174                         if (unlikely(err)) {
1175                                 sci->sc_stage.dirty_file_ptr =
1176                                         list_entry(ii->i_dirty.prev,
1177                                                    struct nilfs_inode_info,
1178                                                    i_dirty);
1179                                 goto break_or_fail;
1180                         }
1181                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1182                         /* XXX: required ? */
1183                 }
1184                 sci->sc_stage.dirty_file_ptr = NULL;
1185                 if (mode == SC_FLUSH_FILE) {
1186                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1187                         return 0;
1188                 }
1189                 nilfs_sc_cstage_inc(sci);
1190                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1191                 fallthrough;
1192         case NILFS_ST_IFILE:
1193                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1194                                               &nilfs_sc_file_ops);
1195                 if (unlikely(err))
1196                         break;
1197                 nilfs_sc_cstage_inc(sci);
1198                 /* Creating a checkpoint */
1199                 err = nilfs_segctor_create_checkpoint(sci);
1200                 if (unlikely(err))
1201                         break;
1202                 fallthrough;
1203         case NILFS_ST_CPFILE:
1204                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1205                                               &nilfs_sc_file_ops);
1206                 if (unlikely(err))
1207                         break;
1208                 nilfs_sc_cstage_inc(sci);
1209                 fallthrough;
1210         case NILFS_ST_SUFILE:
1211                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1212                                          sci->sc_nfreesegs, &ndone);
1213                 if (unlikely(err)) {
1214                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1215                                                   sci->sc_freesegs, ndone,
1216                                                   NULL);
1217                         break;
1218                 }
1219                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1220
1221                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1222                                               &nilfs_sc_file_ops);
1223                 if (unlikely(err))
1224                         break;
1225                 nilfs_sc_cstage_inc(sci);
1226                 fallthrough;
1227         case NILFS_ST_DAT:
1228  dat_stage:
1229                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1230                                               &nilfs_sc_dat_ops);
1231                 if (unlikely(err))
1232                         break;
1233                 if (mode == SC_FLUSH_DAT) {
1234                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1235                         return 0;
1236                 }
1237                 nilfs_sc_cstage_inc(sci);
1238                 fallthrough;
1239         case NILFS_ST_SR:
1240                 if (mode == SC_LSEG_SR) {
1241                         /* Appending a super root */
1242                         err = nilfs_segctor_add_super_root(sci);
1243                         if (unlikely(err))
1244                                 break;
1245                 }
1246                 /* End of a logical segment */
1247                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1248                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1249                 return 0;
1250         case NILFS_ST_DSYNC:
1251  dsync_mode:
1252                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1253                 ii = sci->sc_dsync_inode;
1254                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1255                         break;
1256
1257                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1258                 if (unlikely(err))
1259                         break;
1260                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1261                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1262                 return 0;
1263         case NILFS_ST_DONE:
1264                 return 0;
1265         default:
1266                 BUG();
1267         }
1268
1269  break_or_fail:
1270         return err;
1271 }
1272
1273 /**
1274  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1275  * @sci: nilfs_sc_info
1276  * @nilfs: nilfs object
1277  */
1278 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1279                                             struct the_nilfs *nilfs)
1280 {
1281         struct nilfs_segment_buffer *segbuf, *prev;
1282         __u64 nextnum;
1283         int err, alloc = 0;
1284
1285         segbuf = nilfs_segbuf_new(sci->sc_super);
1286         if (unlikely(!segbuf))
1287                 return -ENOMEM;
1288
1289         if (list_empty(&sci->sc_write_logs)) {
1290                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1291                                  nilfs->ns_pseg_offset, nilfs);
1292                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1293                         nilfs_shift_to_next_segment(nilfs);
1294                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1295                 }
1296
1297                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1298                 nextnum = nilfs->ns_nextnum;
1299
1300                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1301                         /* Start from the head of a new full segment */
1302                         alloc++;
1303         } else {
1304                 /* Continue logs */
1305                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1306                 nilfs_segbuf_map_cont(segbuf, prev);
1307                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1308                 nextnum = prev->sb_nextnum;
1309
1310                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1311                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1312                         segbuf->sb_sum.seg_seq++;
1313                         alloc++;
1314                 }
1315         }
1316
1317         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1318         if (err)
1319                 goto failed;
1320
1321         if (alloc) {
1322                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1323                 if (err)
1324                         goto failed;
1325         }
1326         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1327
1328         BUG_ON(!list_empty(&sci->sc_segbufs));
1329         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1330         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1331         return 0;
1332
1333  failed:
1334         nilfs_segbuf_free(segbuf);
1335         return err;
1336 }
1337
1338 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1339                                          struct the_nilfs *nilfs, int nadd)
1340 {
1341         struct nilfs_segment_buffer *segbuf, *prev;
1342         struct inode *sufile = nilfs->ns_sufile;
1343         __u64 nextnextnum;
1344         LIST_HEAD(list);
1345         int err, ret, i;
1346
1347         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1348         /*
1349          * Since the segment specified with nextnum might be allocated during
1350          * the previous construction, the buffer including its segusage may
1351          * not be dirty.  The following call ensures that the buffer is dirty
1352          * and will pin the buffer on memory until the sufile is written.
1353          */
1354         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1355         if (unlikely(err))
1356                 return err;
1357
1358         for (i = 0; i < nadd; i++) {
1359                 /* extend segment info */
1360                 err = -ENOMEM;
1361                 segbuf = nilfs_segbuf_new(sci->sc_super);
1362                 if (unlikely(!segbuf))
1363                         goto failed;
1364
1365                 /* map this buffer to region of segment on-disk */
1366                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1367                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1368
1369                 /* allocate the next next full segment */
1370                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1371                 if (unlikely(err))
1372                         goto failed_segbuf;
1373
1374                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1375                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1376
1377                 list_add_tail(&segbuf->sb_list, &list);
1378                 prev = segbuf;
1379         }
1380         list_splice_tail(&list, &sci->sc_segbufs);
1381         return 0;
1382
1383  failed_segbuf:
1384         nilfs_segbuf_free(segbuf);
1385  failed:
1386         list_for_each_entry(segbuf, &list, sb_list) {
1387                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1388                 WARN_ON(ret); /* never fails */
1389         }
1390         nilfs_destroy_logs(&list);
1391         return err;
1392 }
1393
1394 static void nilfs_free_incomplete_logs(struct list_head *logs,
1395                                        struct the_nilfs *nilfs)
1396 {
1397         struct nilfs_segment_buffer *segbuf, *prev;
1398         struct inode *sufile = nilfs->ns_sufile;
1399         int ret;
1400
1401         segbuf = NILFS_FIRST_SEGBUF(logs);
1402         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1403                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1404                 WARN_ON(ret); /* never fails */
1405         }
1406         if (atomic_read(&segbuf->sb_err)) {
1407                 /* Case 1: The first segment failed */
1408                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1409                         /*
1410                          * Case 1a:  Partial segment appended into an existing
1411                          * segment
1412                          */
1413                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1414                                                 segbuf->sb_fseg_end);
1415                 else /* Case 1b:  New full segment */
1416                         set_nilfs_discontinued(nilfs);
1417         }
1418
1419         prev = segbuf;
1420         list_for_each_entry_continue(segbuf, logs, sb_list) {
1421                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1422                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1423                         WARN_ON(ret); /* never fails */
1424                 }
1425                 if (atomic_read(&segbuf->sb_err) &&
1426                     segbuf->sb_segnum != nilfs->ns_nextnum)
1427                         /* Case 2: extended segment (!= next) failed */
1428                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1429                 prev = segbuf;
1430         }
1431 }
1432
1433 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1434                                           struct inode *sufile)
1435 {
1436         struct nilfs_segment_buffer *segbuf;
1437         unsigned long live_blocks;
1438         int ret;
1439
1440         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1441                 live_blocks = segbuf->sb_sum.nblocks +
1442                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1443                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1444                                                      live_blocks,
1445                                                      sci->sc_seg_ctime);
1446                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1447         }
1448 }
1449
1450 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1451 {
1452         struct nilfs_segment_buffer *segbuf;
1453         int ret;
1454
1455         segbuf = NILFS_FIRST_SEGBUF(logs);
1456         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1457                                              segbuf->sb_pseg_start -
1458                                              segbuf->sb_fseg_start, 0);
1459         WARN_ON(ret); /* always succeed because the segusage is dirty */
1460
1461         list_for_each_entry_continue(segbuf, logs, sb_list) {
1462                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1463                                                      0, 0);
1464                 WARN_ON(ret); /* always succeed */
1465         }
1466 }
1467
1468 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1469                                             struct nilfs_segment_buffer *last,
1470                                             struct inode *sufile)
1471 {
1472         struct nilfs_segment_buffer *segbuf = last;
1473         int ret;
1474
1475         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1476                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1477                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1478                 WARN_ON(ret);
1479         }
1480         nilfs_truncate_logs(&sci->sc_segbufs, last);
1481 }
1482
1483
1484 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1485                                  struct the_nilfs *nilfs, int mode)
1486 {
1487         struct nilfs_cstage prev_stage = sci->sc_stage;
1488         int err, nadd = 1;
1489
1490         /* Collection retry loop */
1491         for (;;) {
1492                 sci->sc_nblk_this_inc = 0;
1493                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1494
1495                 err = nilfs_segctor_reset_segment_buffer(sci);
1496                 if (unlikely(err))
1497                         goto failed;
1498
1499                 err = nilfs_segctor_collect_blocks(sci, mode);
1500                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1501                 if (!err)
1502                         break;
1503
1504                 if (unlikely(err != -E2BIG))
1505                         goto failed;
1506
1507                 /* The current segment is filled up */
1508                 if (mode != SC_LSEG_SR ||
1509                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1510                         break;
1511
1512                 nilfs_clear_logs(&sci->sc_segbufs);
1513
1514                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1515                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1516                                                         sci->sc_freesegs,
1517                                                         sci->sc_nfreesegs,
1518                                                         NULL);
1519                         WARN_ON(err); /* do not happen */
1520                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1521                 }
1522
1523                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1524                 if (unlikely(err))
1525                         return err;
1526
1527                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1528                 sci->sc_stage = prev_stage;
1529         }
1530         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1531         return 0;
1532
1533  failed:
1534         return err;
1535 }
1536
1537 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1538                                       struct buffer_head *new_bh)
1539 {
1540         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1541
1542         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1543         /* The caller must release old_bh */
1544 }
1545
1546 static int
1547 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1548                                      struct nilfs_segment_buffer *segbuf,
1549                                      int mode)
1550 {
1551         struct inode *inode = NULL;
1552         sector_t blocknr;
1553         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1554         unsigned long nblocks = 0, ndatablk = 0;
1555         const struct nilfs_sc_operations *sc_op = NULL;
1556         struct nilfs_segsum_pointer ssp;
1557         struct nilfs_finfo *finfo = NULL;
1558         union nilfs_binfo binfo;
1559         struct buffer_head *bh, *bh_org;
1560         ino_t ino = 0;
1561         int err = 0;
1562
1563         if (!nfinfo)
1564                 goto out;
1565
1566         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1567         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1568         ssp.offset = sizeof(struct nilfs_segment_summary);
1569
1570         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1571                 if (bh == segbuf->sb_super_root)
1572                         break;
1573                 if (!finfo) {
1574                         finfo = nilfs_segctor_map_segsum_entry(
1575                                 sci, &ssp, sizeof(*finfo));
1576                         ino = le64_to_cpu(finfo->fi_ino);
1577                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1578                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1579
1580                         inode = bh->b_page->mapping->host;
1581
1582                         if (mode == SC_LSEG_DSYNC)
1583                                 sc_op = &nilfs_sc_dsync_ops;
1584                         else if (ino == NILFS_DAT_INO)
1585                                 sc_op = &nilfs_sc_dat_ops;
1586                         else /* file blocks */
1587                                 sc_op = &nilfs_sc_file_ops;
1588                 }
1589                 bh_org = bh;
1590                 get_bh(bh_org);
1591                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1592                                         &binfo);
1593                 if (bh != bh_org)
1594                         nilfs_list_replace_buffer(bh_org, bh);
1595                 brelse(bh_org);
1596                 if (unlikely(err))
1597                         goto failed_bmap;
1598
1599                 if (ndatablk > 0)
1600                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1601                 else
1602                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1603
1604                 blocknr++;
1605                 if (--nblocks == 0) {
1606                         finfo = NULL;
1607                         if (--nfinfo == 0)
1608                                 break;
1609                 } else if (ndatablk > 0)
1610                         ndatablk--;
1611         }
1612  out:
1613         return 0;
1614
1615  failed_bmap:
1616         return err;
1617 }
1618
1619 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1620 {
1621         struct nilfs_segment_buffer *segbuf;
1622         int err;
1623
1624         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1625                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1626                 if (unlikely(err))
1627                         return err;
1628                 nilfs_segbuf_fill_in_segsum(segbuf);
1629         }
1630         return 0;
1631 }
1632
1633 static void nilfs_begin_page_io(struct page *page)
1634 {
1635         if (!page || PageWriteback(page))
1636                 /*
1637                  * For split b-tree node pages, this function may be called
1638                  * twice.  We ignore the 2nd or later calls by this check.
1639                  */
1640                 return;
1641
1642         lock_page(page);
1643         clear_page_dirty_for_io(page);
1644         set_page_writeback(page);
1645         unlock_page(page);
1646 }
1647
1648 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1649 {
1650         struct nilfs_segment_buffer *segbuf;
1651         struct page *bd_page = NULL, *fs_page = NULL;
1652
1653         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1654                 struct buffer_head *bh;
1655
1656                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1657                                     b_assoc_buffers) {
1658                         if (bh->b_page != bd_page) {
1659                                 if (bd_page) {
1660                                         lock_page(bd_page);
1661                                         clear_page_dirty_for_io(bd_page);
1662                                         set_page_writeback(bd_page);
1663                                         unlock_page(bd_page);
1664                                 }
1665                                 bd_page = bh->b_page;
1666                         }
1667                 }
1668
1669                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1670                                     b_assoc_buffers) {
1671                         set_buffer_async_write(bh);
1672                         if (bh == segbuf->sb_super_root) {
1673                                 if (bh->b_page != bd_page) {
1674                                         lock_page(bd_page);
1675                                         clear_page_dirty_for_io(bd_page);
1676                                         set_page_writeback(bd_page);
1677                                         unlock_page(bd_page);
1678                                         bd_page = bh->b_page;
1679                                 }
1680                                 break;
1681                         }
1682                         if (bh->b_page != fs_page) {
1683                                 nilfs_begin_page_io(fs_page);
1684                                 fs_page = bh->b_page;
1685                         }
1686                 }
1687         }
1688         if (bd_page) {
1689                 lock_page(bd_page);
1690                 clear_page_dirty_for_io(bd_page);
1691                 set_page_writeback(bd_page);
1692                 unlock_page(bd_page);
1693         }
1694         nilfs_begin_page_io(fs_page);
1695 }
1696
1697 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1698                                struct the_nilfs *nilfs)
1699 {
1700         int ret;
1701
1702         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1703         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1704         return ret;
1705 }
1706
1707 static void nilfs_end_page_io(struct page *page, int err)
1708 {
1709         if (!page)
1710                 return;
1711
1712         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1713                 /*
1714                  * For b-tree node pages, this function may be called twice
1715                  * or more because they might be split in a segment.
1716                  */
1717                 if (PageDirty(page)) {
1718                         /*
1719                          * For pages holding split b-tree node buffers, dirty
1720                          * flag on the buffers may be cleared discretely.
1721                          * In that case, the page is once redirtied for
1722                          * remaining buffers, and it must be cancelled if
1723                          * all the buffers get cleaned later.
1724                          */
1725                         lock_page(page);
1726                         if (nilfs_page_buffers_clean(page))
1727                                 __nilfs_clear_page_dirty(page);
1728                         unlock_page(page);
1729                 }
1730                 return;
1731         }
1732
1733         if (!err) {
1734                 if (!nilfs_page_buffers_clean(page))
1735                         __set_page_dirty_nobuffers(page);
1736                 ClearPageError(page);
1737         } else {
1738                 __set_page_dirty_nobuffers(page);
1739                 SetPageError(page);
1740         }
1741
1742         end_page_writeback(page);
1743 }
1744
1745 static void nilfs_abort_logs(struct list_head *logs, int err)
1746 {
1747         struct nilfs_segment_buffer *segbuf;
1748         struct page *bd_page = NULL, *fs_page = NULL;
1749         struct buffer_head *bh;
1750
1751         if (list_empty(logs))
1752                 return;
1753
1754         list_for_each_entry(segbuf, logs, sb_list) {
1755                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1756                                     b_assoc_buffers) {
1757                         if (bh->b_page != bd_page) {
1758                                 if (bd_page)
1759                                         end_page_writeback(bd_page);
1760                                 bd_page = bh->b_page;
1761                         }
1762                 }
1763
1764                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1765                                     b_assoc_buffers) {
1766                         clear_buffer_async_write(bh);
1767                         if (bh == segbuf->sb_super_root) {
1768                                 if (bh->b_page != bd_page) {
1769                                         end_page_writeback(bd_page);
1770                                         bd_page = bh->b_page;
1771                                 }
1772                                 break;
1773                         }
1774                         if (bh->b_page != fs_page) {
1775                                 nilfs_end_page_io(fs_page, err);
1776                                 fs_page = bh->b_page;
1777                         }
1778                 }
1779         }
1780         if (bd_page)
1781                 end_page_writeback(bd_page);
1782
1783         nilfs_end_page_io(fs_page, err);
1784 }
1785
1786 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1787                                              struct the_nilfs *nilfs, int err)
1788 {
1789         LIST_HEAD(logs);
1790         int ret;
1791
1792         list_splice_tail_init(&sci->sc_write_logs, &logs);
1793         ret = nilfs_wait_on_logs(&logs);
1794         nilfs_abort_logs(&logs, ret ? : err);
1795
1796         list_splice_tail_init(&sci->sc_segbufs, &logs);
1797         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1798         nilfs_free_incomplete_logs(&logs, nilfs);
1799
1800         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1801                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1802                                                 sci->sc_freesegs,
1803                                                 sci->sc_nfreesegs,
1804                                                 NULL);
1805                 WARN_ON(ret); /* do not happen */
1806         }
1807
1808         nilfs_destroy_logs(&logs);
1809 }
1810
1811 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1812                                    struct nilfs_segment_buffer *segbuf)
1813 {
1814         nilfs->ns_segnum = segbuf->sb_segnum;
1815         nilfs->ns_nextnum = segbuf->sb_nextnum;
1816         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1817                 + segbuf->sb_sum.nblocks;
1818         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1819         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1820 }
1821
1822 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1823 {
1824         struct nilfs_segment_buffer *segbuf;
1825         struct page *bd_page = NULL, *fs_page = NULL;
1826         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1827         int update_sr = false;
1828
1829         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1830                 struct buffer_head *bh;
1831
1832                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1833                                     b_assoc_buffers) {
1834                         set_buffer_uptodate(bh);
1835                         clear_buffer_dirty(bh);
1836                         if (bh->b_page != bd_page) {
1837                                 if (bd_page)
1838                                         end_page_writeback(bd_page);
1839                                 bd_page = bh->b_page;
1840                         }
1841                 }
1842                 /*
1843                  * We assume that the buffers which belong to the same page
1844                  * continue over the buffer list.
1845                  * Under this assumption, the last BHs of pages is
1846                  * identifiable by the discontinuity of bh->b_page
1847                  * (page != fs_page).
1848                  *
1849                  * For B-tree node blocks, however, this assumption is not
1850                  * guaranteed.  The cleanup code of B-tree node pages needs
1851                  * special care.
1852                  */
1853                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1854                                     b_assoc_buffers) {
1855                         const unsigned long set_bits = BIT(BH_Uptodate);
1856                         const unsigned long clear_bits =
1857                                 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1858                                  BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1859                                  BIT(BH_NILFS_Redirected));
1860
1861                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1862                         if (bh == segbuf->sb_super_root) {
1863                                 if (bh->b_page != bd_page) {
1864                                         end_page_writeback(bd_page);
1865                                         bd_page = bh->b_page;
1866                                 }
1867                                 update_sr = true;
1868                                 break;
1869                         }
1870                         if (bh->b_page != fs_page) {
1871                                 nilfs_end_page_io(fs_page, 0);
1872                                 fs_page = bh->b_page;
1873                         }
1874                 }
1875
1876                 if (!nilfs_segbuf_simplex(segbuf)) {
1877                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1878                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1879                                 sci->sc_lseg_stime = jiffies;
1880                         }
1881                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1882                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1883                 }
1884         }
1885         /*
1886          * Since pages may continue over multiple segment buffers,
1887          * end of the last page must be checked outside of the loop.
1888          */
1889         if (bd_page)
1890                 end_page_writeback(bd_page);
1891
1892         nilfs_end_page_io(fs_page, 0);
1893
1894         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1895
1896         if (nilfs_doing_gc())
1897                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1898         else
1899                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1900
1901         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1902
1903         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1904         nilfs_set_next_segment(nilfs, segbuf);
1905
1906         if (update_sr) {
1907                 nilfs->ns_flushed_device = 0;
1908                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1909                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1910
1911                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1912                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1913                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1914                 nilfs_segctor_clear_metadata_dirty(sci);
1915         } else
1916                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1917 }
1918
1919 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1920 {
1921         int ret;
1922
1923         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1924         if (!ret) {
1925                 nilfs_segctor_complete_write(sci);
1926                 nilfs_destroy_logs(&sci->sc_write_logs);
1927         }
1928         return ret;
1929 }
1930
1931 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1932                                              struct the_nilfs *nilfs)
1933 {
1934         struct nilfs_inode_info *ii, *n;
1935         struct inode *ifile = sci->sc_root->ifile;
1936
1937         spin_lock(&nilfs->ns_inode_lock);
1938  retry:
1939         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1940                 if (!ii->i_bh) {
1941                         struct buffer_head *ibh;
1942                         int err;
1943
1944                         spin_unlock(&nilfs->ns_inode_lock);
1945                         err = nilfs_ifile_get_inode_block(
1946                                 ifile, ii->vfs_inode.i_ino, &ibh);
1947                         if (unlikely(err)) {
1948                                 nilfs_warn(sci->sc_super,
1949                                            "log writer: error %d getting inode block (ino=%lu)",
1950                                            err, ii->vfs_inode.i_ino);
1951                                 return err;
1952                         }
1953                         spin_lock(&nilfs->ns_inode_lock);
1954                         if (likely(!ii->i_bh))
1955                                 ii->i_bh = ibh;
1956                         else
1957                                 brelse(ibh);
1958                         goto retry;
1959                 }
1960
1961                 // Always redirty the buffer to avoid race condition
1962                 mark_buffer_dirty(ii->i_bh);
1963                 nilfs_mdt_mark_dirty(ifile);
1964
1965                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1966                 set_bit(NILFS_I_BUSY, &ii->i_state);
1967                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1968         }
1969         spin_unlock(&nilfs->ns_inode_lock);
1970
1971         return 0;
1972 }
1973
1974 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1975                                              struct the_nilfs *nilfs)
1976 {
1977         struct nilfs_inode_info *ii, *n;
1978         int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1979         int defer_iput = false;
1980
1981         spin_lock(&nilfs->ns_inode_lock);
1982         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1983                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1984                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1985                         continue;
1986
1987                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1988                 brelse(ii->i_bh);
1989                 ii->i_bh = NULL;
1990                 list_del_init(&ii->i_dirty);
1991                 if (!ii->vfs_inode.i_nlink || during_mount) {
1992                         /*
1993                          * Defer calling iput() to avoid deadlocks if
1994                          * i_nlink == 0 or mount is not yet finished.
1995                          */
1996                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
1997                         defer_iput = true;
1998                 } else {
1999                         spin_unlock(&nilfs->ns_inode_lock);
2000                         iput(&ii->vfs_inode);
2001                         spin_lock(&nilfs->ns_inode_lock);
2002                 }
2003         }
2004         spin_unlock(&nilfs->ns_inode_lock);
2005
2006         if (defer_iput)
2007                 schedule_work(&sci->sc_iput_work);
2008 }
2009
2010 /*
2011  * Main procedure of segment constructor
2012  */
2013 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2014 {
2015         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2016         int err;
2017
2018         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2019         sci->sc_cno = nilfs->ns_cno;
2020
2021         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2022         if (unlikely(err))
2023                 goto out;
2024
2025         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2026                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2027
2028         if (nilfs_segctor_clean(sci))
2029                 goto out;
2030
2031         do {
2032                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2033
2034                 err = nilfs_segctor_begin_construction(sci, nilfs);
2035                 if (unlikely(err))
2036                         goto out;
2037
2038                 /* Update time stamp */
2039                 sci->sc_seg_ctime = ktime_get_real_seconds();
2040
2041                 err = nilfs_segctor_collect(sci, nilfs, mode);
2042                 if (unlikely(err))
2043                         goto failed;
2044
2045                 /* Avoid empty segment */
2046                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2047                     nilfs_segbuf_empty(sci->sc_curseg)) {
2048                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2049                         goto out;
2050                 }
2051
2052                 err = nilfs_segctor_assign(sci, mode);
2053                 if (unlikely(err))
2054                         goto failed;
2055
2056                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2057                         nilfs_segctor_fill_in_file_bmap(sci);
2058
2059                 if (mode == SC_LSEG_SR &&
2060                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2061                         err = nilfs_segctor_fill_in_checkpoint(sci);
2062                         if (unlikely(err))
2063                                 goto failed_to_write;
2064
2065                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2066                 }
2067                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2068
2069                 /* Write partial segments */
2070                 nilfs_segctor_prepare_write(sci);
2071
2072                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2073                                             nilfs->ns_crc_seed);
2074
2075                 err = nilfs_segctor_write(sci, nilfs);
2076                 if (unlikely(err))
2077                         goto failed_to_write;
2078
2079                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2080                     nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2081                         /*
2082                          * At this point, we avoid double buffering
2083                          * for blocksize < pagesize because page dirty
2084                          * flag is turned off during write and dirty
2085                          * buffers are not properly collected for
2086                          * pages crossing over segments.
2087                          */
2088                         err = nilfs_segctor_wait(sci);
2089                         if (err)
2090                                 goto failed_to_write;
2091                 }
2092         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2093
2094  out:
2095         nilfs_segctor_drop_written_files(sci, nilfs);
2096         return err;
2097
2098  failed_to_write:
2099         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2100                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2101
2102  failed:
2103         if (nilfs_doing_gc())
2104                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2105         nilfs_segctor_abort_construction(sci, nilfs, err);
2106         goto out;
2107 }
2108
2109 /**
2110  * nilfs_segctor_start_timer - set timer of background write
2111  * @sci: nilfs_sc_info
2112  *
2113  * If the timer has already been set, it ignores the new request.
2114  * This function MUST be called within a section locking the segment
2115  * semaphore.
2116  */
2117 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2118 {
2119         spin_lock(&sci->sc_state_lock);
2120         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2121                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2122                 add_timer(&sci->sc_timer);
2123                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2124         }
2125         spin_unlock(&sci->sc_state_lock);
2126 }
2127
2128 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2129 {
2130         spin_lock(&sci->sc_state_lock);
2131         if (!(sci->sc_flush_request & BIT(bn))) {
2132                 unsigned long prev_req = sci->sc_flush_request;
2133
2134                 sci->sc_flush_request |= BIT(bn);
2135                 if (!prev_req)
2136                         wake_up(&sci->sc_wait_daemon);
2137         }
2138         spin_unlock(&sci->sc_state_lock);
2139 }
2140
2141 /**
2142  * nilfs_flush_segment - trigger a segment construction for resource control
2143  * @sb: super block
2144  * @ino: inode number of the file to be flushed out.
2145  */
2146 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2147 {
2148         struct the_nilfs *nilfs = sb->s_fs_info;
2149         struct nilfs_sc_info *sci = nilfs->ns_writer;
2150
2151         if (!sci || nilfs_doing_construction())
2152                 return;
2153         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2154                                         /* assign bit 0 to data files */
2155 }
2156
2157 struct nilfs_segctor_wait_request {
2158         wait_queue_entry_t      wq;
2159         __u32           seq;
2160         int             err;
2161         atomic_t        done;
2162 };
2163
2164 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2165 {
2166         struct nilfs_segctor_wait_request wait_req;
2167         int err = 0;
2168
2169         spin_lock(&sci->sc_state_lock);
2170         init_wait(&wait_req.wq);
2171         wait_req.err = 0;
2172         atomic_set(&wait_req.done, 0);
2173         wait_req.seq = ++sci->sc_seq_request;
2174         spin_unlock(&sci->sc_state_lock);
2175
2176         init_waitqueue_entry(&wait_req.wq, current);
2177         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2178         set_current_state(TASK_INTERRUPTIBLE);
2179         wake_up(&sci->sc_wait_daemon);
2180
2181         for (;;) {
2182                 if (atomic_read(&wait_req.done)) {
2183                         err = wait_req.err;
2184                         break;
2185                 }
2186                 if (!signal_pending(current)) {
2187                         schedule();
2188                         continue;
2189                 }
2190                 err = -ERESTARTSYS;
2191                 break;
2192         }
2193         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2194         return err;
2195 }
2196
2197 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2198 {
2199         struct nilfs_segctor_wait_request *wrq, *n;
2200         unsigned long flags;
2201
2202         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2203         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2204                 if (!atomic_read(&wrq->done) &&
2205                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2206                         wrq->err = err;
2207                         atomic_set(&wrq->done, 1);
2208                 }
2209                 if (atomic_read(&wrq->done)) {
2210                         wrq->wq.func(&wrq->wq,
2211                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2212                                      0, NULL);
2213                 }
2214         }
2215         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2216 }
2217
2218 /**
2219  * nilfs_construct_segment - construct a logical segment
2220  * @sb: super block
2221  *
2222  * Return Value: On success, 0 is retured. On errors, one of the following
2223  * negative error code is returned.
2224  *
2225  * %-EROFS - Read only filesystem.
2226  *
2227  * %-EIO - I/O error
2228  *
2229  * %-ENOSPC - No space left on device (only in a panic state).
2230  *
2231  * %-ERESTARTSYS - Interrupted.
2232  *
2233  * %-ENOMEM - Insufficient memory available.
2234  */
2235 int nilfs_construct_segment(struct super_block *sb)
2236 {
2237         struct the_nilfs *nilfs = sb->s_fs_info;
2238         struct nilfs_sc_info *sci = nilfs->ns_writer;
2239         struct nilfs_transaction_info *ti;
2240         int err;
2241
2242         if (!sci)
2243                 return -EROFS;
2244
2245         /* A call inside transactions causes a deadlock. */
2246         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2247
2248         err = nilfs_segctor_sync(sci);
2249         return err;
2250 }
2251
2252 /**
2253  * nilfs_construct_dsync_segment - construct a data-only logical segment
2254  * @sb: super block
2255  * @inode: inode whose data blocks should be written out
2256  * @start: start byte offset
2257  * @end: end byte offset (inclusive)
2258  *
2259  * Return Value: On success, 0 is retured. On errors, one of the following
2260  * negative error code is returned.
2261  *
2262  * %-EROFS - Read only filesystem.
2263  *
2264  * %-EIO - I/O error
2265  *
2266  * %-ENOSPC - No space left on device (only in a panic state).
2267  *
2268  * %-ERESTARTSYS - Interrupted.
2269  *
2270  * %-ENOMEM - Insufficient memory available.
2271  */
2272 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2273                                   loff_t start, loff_t end)
2274 {
2275         struct the_nilfs *nilfs = sb->s_fs_info;
2276         struct nilfs_sc_info *sci = nilfs->ns_writer;
2277         struct nilfs_inode_info *ii;
2278         struct nilfs_transaction_info ti;
2279         int err = 0;
2280
2281         if (!sci)
2282                 return -EROFS;
2283
2284         nilfs_transaction_lock(sb, &ti, 0);
2285
2286         ii = NILFS_I(inode);
2287         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2288             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2289             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2290             nilfs_discontinued(nilfs)) {
2291                 nilfs_transaction_unlock(sb);
2292                 err = nilfs_segctor_sync(sci);
2293                 return err;
2294         }
2295
2296         spin_lock(&nilfs->ns_inode_lock);
2297         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2298             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2299                 spin_unlock(&nilfs->ns_inode_lock);
2300                 nilfs_transaction_unlock(sb);
2301                 return 0;
2302         }
2303         spin_unlock(&nilfs->ns_inode_lock);
2304         sci->sc_dsync_inode = ii;
2305         sci->sc_dsync_start = start;
2306         sci->sc_dsync_end = end;
2307
2308         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2309         if (!err)
2310                 nilfs->ns_flushed_device = 0;
2311
2312         nilfs_transaction_unlock(sb);
2313         return err;
2314 }
2315
2316 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2317 #define FLUSH_DAT_BIT   BIT(NILFS_DAT_INO) /* DAT only */
2318
2319 /**
2320  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2321  * @sci: segment constructor object
2322  */
2323 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2324 {
2325         spin_lock(&sci->sc_state_lock);
2326         sci->sc_seq_accepted = sci->sc_seq_request;
2327         spin_unlock(&sci->sc_state_lock);
2328         del_timer_sync(&sci->sc_timer);
2329 }
2330
2331 /**
2332  * nilfs_segctor_notify - notify the result of request to caller threads
2333  * @sci: segment constructor object
2334  * @mode: mode of log forming
2335  * @err: error code to be notified
2336  */
2337 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2338 {
2339         /* Clear requests (even when the construction failed) */
2340         spin_lock(&sci->sc_state_lock);
2341
2342         if (mode == SC_LSEG_SR) {
2343                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2344                 sci->sc_seq_done = sci->sc_seq_accepted;
2345                 nilfs_segctor_wakeup(sci, err);
2346                 sci->sc_flush_request = 0;
2347         } else {
2348                 if (mode == SC_FLUSH_FILE)
2349                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2350                 else if (mode == SC_FLUSH_DAT)
2351                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2352
2353                 /* re-enable timer if checkpoint creation was not done */
2354                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2355                     time_before(jiffies, sci->sc_timer.expires))
2356                         add_timer(&sci->sc_timer);
2357         }
2358         spin_unlock(&sci->sc_state_lock);
2359 }
2360
2361 /**
2362  * nilfs_segctor_construct - form logs and write them to disk
2363  * @sci: segment constructor object
2364  * @mode: mode of log forming
2365  */
2366 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2367 {
2368         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2369         struct nilfs_super_block **sbp;
2370         int err = 0;
2371
2372         nilfs_segctor_accept(sci);
2373
2374         if (nilfs_discontinued(nilfs))
2375                 mode = SC_LSEG_SR;
2376         if (!nilfs_segctor_confirm(sci))
2377                 err = nilfs_segctor_do_construct(sci, mode);
2378
2379         if (likely(!err)) {
2380                 if (mode != SC_FLUSH_DAT)
2381                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2382                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2383                     nilfs_discontinued(nilfs)) {
2384                         down_write(&nilfs->ns_sem);
2385                         err = -EIO;
2386                         sbp = nilfs_prepare_super(sci->sc_super,
2387                                                   nilfs_sb_will_flip(nilfs));
2388                         if (likely(sbp)) {
2389                                 nilfs_set_log_cursor(sbp[0], nilfs);
2390                                 err = nilfs_commit_super(sci->sc_super,
2391                                                          NILFS_SB_COMMIT);
2392                         }
2393                         up_write(&nilfs->ns_sem);
2394                 }
2395         }
2396
2397         nilfs_segctor_notify(sci, mode, err);
2398         return err;
2399 }
2400
2401 static void nilfs_construction_timeout(struct timer_list *t)
2402 {
2403         struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2404
2405         wake_up_process(sci->sc_timer_task);
2406 }
2407
2408 static void
2409 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2410 {
2411         struct nilfs_inode_info *ii, *n;
2412
2413         list_for_each_entry_safe(ii, n, head, i_dirty) {
2414                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2415                         continue;
2416                 list_del_init(&ii->i_dirty);
2417                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2418                 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2419                 iput(&ii->vfs_inode);
2420         }
2421 }
2422
2423 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2424                          void **kbufs)
2425 {
2426         struct the_nilfs *nilfs = sb->s_fs_info;
2427         struct nilfs_sc_info *sci = nilfs->ns_writer;
2428         struct nilfs_transaction_info ti;
2429         int err;
2430
2431         if (unlikely(!sci))
2432                 return -EROFS;
2433
2434         nilfs_transaction_lock(sb, &ti, 1);
2435
2436         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2437         if (unlikely(err))
2438                 goto out_unlock;
2439
2440         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2441         if (unlikely(err)) {
2442                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2443                 goto out_unlock;
2444         }
2445
2446         sci->sc_freesegs = kbufs[4];
2447         sci->sc_nfreesegs = argv[4].v_nmembs;
2448         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2449
2450         for (;;) {
2451                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2452                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2453
2454                 if (likely(!err))
2455                         break;
2456
2457                 nilfs_warn(sb, "error %d cleaning segments", err);
2458                 set_current_state(TASK_INTERRUPTIBLE);
2459                 schedule_timeout(sci->sc_interval);
2460         }
2461         if (nilfs_test_opt(nilfs, DISCARD)) {
2462                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2463                                                  sci->sc_nfreesegs);
2464                 if (ret) {
2465                         nilfs_warn(sb,
2466                                    "error %d on discard request, turning discards off for the device",
2467                                    ret);
2468                         nilfs_clear_opt(nilfs, DISCARD);
2469                 }
2470         }
2471
2472  out_unlock:
2473         sci->sc_freesegs = NULL;
2474         sci->sc_nfreesegs = 0;
2475         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2476         nilfs_transaction_unlock(sb);
2477         return err;
2478 }
2479
2480 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2481 {
2482         struct nilfs_transaction_info ti;
2483
2484         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2485         nilfs_segctor_construct(sci, mode);
2486
2487         /*
2488          * Unclosed segment should be retried.  We do this using sc_timer.
2489          * Timeout of sc_timer will invoke complete construction which leads
2490          * to close the current logical segment.
2491          */
2492         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2493                 nilfs_segctor_start_timer(sci);
2494
2495         nilfs_transaction_unlock(sci->sc_super);
2496 }
2497
2498 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2499 {
2500         int mode = 0;
2501
2502         spin_lock(&sci->sc_state_lock);
2503         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2504                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2505         spin_unlock(&sci->sc_state_lock);
2506
2507         if (mode) {
2508                 nilfs_segctor_do_construct(sci, mode);
2509
2510                 spin_lock(&sci->sc_state_lock);
2511                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2512                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2513                 spin_unlock(&sci->sc_state_lock);
2514         }
2515         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2516 }
2517
2518 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2519 {
2520         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2521             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2522                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2523                         return SC_FLUSH_FILE;
2524                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2525                         return SC_FLUSH_DAT;
2526         }
2527         return SC_LSEG_SR;
2528 }
2529
2530 /**
2531  * nilfs_segctor_thread - main loop of the segment constructor thread.
2532  * @arg: pointer to a struct nilfs_sc_info.
2533  *
2534  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2535  * to execute segment constructions.
2536  */
2537 static int nilfs_segctor_thread(void *arg)
2538 {
2539         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2540         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2541         int timeout = 0;
2542
2543         sci->sc_timer_task = current;
2544
2545         /* start sync. */
2546         sci->sc_task = current;
2547         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2548         nilfs_info(sci->sc_super,
2549                    "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2550                    sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2551
2552         spin_lock(&sci->sc_state_lock);
2553  loop:
2554         for (;;) {
2555                 int mode;
2556
2557                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2558                         goto end_thread;
2559
2560                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2561                         mode = SC_LSEG_SR;
2562                 else if (sci->sc_flush_request)
2563                         mode = nilfs_segctor_flush_mode(sci);
2564                 else
2565                         break;
2566
2567                 spin_unlock(&sci->sc_state_lock);
2568                 nilfs_segctor_thread_construct(sci, mode);
2569                 spin_lock(&sci->sc_state_lock);
2570                 timeout = 0;
2571         }
2572
2573
2574         if (freezing(current)) {
2575                 spin_unlock(&sci->sc_state_lock);
2576                 try_to_freeze();
2577                 spin_lock(&sci->sc_state_lock);
2578         } else {
2579                 DEFINE_WAIT(wait);
2580                 int should_sleep = 1;
2581
2582                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2583                                 TASK_INTERRUPTIBLE);
2584
2585                 if (sci->sc_seq_request != sci->sc_seq_done)
2586                         should_sleep = 0;
2587                 else if (sci->sc_flush_request)
2588                         should_sleep = 0;
2589                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2590                         should_sleep = time_before(jiffies,
2591                                         sci->sc_timer.expires);
2592
2593                 if (should_sleep) {
2594                         spin_unlock(&sci->sc_state_lock);
2595                         schedule();
2596                         spin_lock(&sci->sc_state_lock);
2597                 }
2598                 finish_wait(&sci->sc_wait_daemon, &wait);
2599                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2600                            time_after_eq(jiffies, sci->sc_timer.expires));
2601
2602                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2603                         set_nilfs_discontinued(nilfs);
2604         }
2605         goto loop;
2606
2607  end_thread:
2608         spin_unlock(&sci->sc_state_lock);
2609
2610         /* end sync. */
2611         sci->sc_task = NULL;
2612         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2613         return 0;
2614 }
2615
2616 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2617 {
2618         struct task_struct *t;
2619
2620         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2621         if (IS_ERR(t)) {
2622                 int err = PTR_ERR(t);
2623
2624                 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2625                           err);
2626                 return err;
2627         }
2628         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2629         return 0;
2630 }
2631
2632 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2633         __acquires(&sci->sc_state_lock)
2634         __releases(&sci->sc_state_lock)
2635 {
2636         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2637
2638         while (sci->sc_task) {
2639                 wake_up(&sci->sc_wait_daemon);
2640                 spin_unlock(&sci->sc_state_lock);
2641                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2642                 spin_lock(&sci->sc_state_lock);
2643         }
2644 }
2645
2646 /*
2647  * Setup & clean-up functions
2648  */
2649 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2650                                                struct nilfs_root *root)
2651 {
2652         struct the_nilfs *nilfs = sb->s_fs_info;
2653         struct nilfs_sc_info *sci;
2654
2655         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2656         if (!sci)
2657                 return NULL;
2658
2659         sci->sc_super = sb;
2660
2661         nilfs_get_root(root);
2662         sci->sc_root = root;
2663
2664         init_waitqueue_head(&sci->sc_wait_request);
2665         init_waitqueue_head(&sci->sc_wait_daemon);
2666         init_waitqueue_head(&sci->sc_wait_task);
2667         spin_lock_init(&sci->sc_state_lock);
2668         INIT_LIST_HEAD(&sci->sc_dirty_files);
2669         INIT_LIST_HEAD(&sci->sc_segbufs);
2670         INIT_LIST_HEAD(&sci->sc_write_logs);
2671         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2672         INIT_LIST_HEAD(&sci->sc_iput_queue);
2673         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2674         timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2675
2676         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2677         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2678         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2679
2680         if (nilfs->ns_interval)
2681                 sci->sc_interval = HZ * nilfs->ns_interval;
2682         if (nilfs->ns_watermark)
2683                 sci->sc_watermark = nilfs->ns_watermark;
2684         return sci;
2685 }
2686
2687 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2688 {
2689         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2690
2691         /*
2692          * The segctord thread was stopped and its timer was removed.
2693          * But some tasks remain.
2694          */
2695         do {
2696                 struct nilfs_transaction_info ti;
2697
2698                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2699                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2700                 nilfs_transaction_unlock(sci->sc_super);
2701
2702                 flush_work(&sci->sc_iput_work);
2703
2704         } while (ret && retrycount-- > 0);
2705 }
2706
2707 /**
2708  * nilfs_segctor_destroy - destroy the segment constructor.
2709  * @sci: nilfs_sc_info
2710  *
2711  * nilfs_segctor_destroy() kills the segctord thread and frees
2712  * the nilfs_sc_info struct.
2713  * Caller must hold the segment semaphore.
2714  */
2715 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2716 {
2717         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2718         int flag;
2719
2720         up_write(&nilfs->ns_segctor_sem);
2721
2722         spin_lock(&sci->sc_state_lock);
2723         nilfs_segctor_kill_thread(sci);
2724         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2725                 || sci->sc_seq_request != sci->sc_seq_done);
2726         spin_unlock(&sci->sc_state_lock);
2727
2728         if (flush_work(&sci->sc_iput_work))
2729                 flag = true;
2730
2731         if (flag || !nilfs_segctor_confirm(sci))
2732                 nilfs_segctor_write_out(sci);
2733
2734         if (!list_empty(&sci->sc_dirty_files)) {
2735                 nilfs_warn(sci->sc_super,
2736                            "disposed unprocessed dirty file(s) when stopping log writer");
2737                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2738         }
2739
2740         if (!list_empty(&sci->sc_iput_queue)) {
2741                 nilfs_warn(sci->sc_super,
2742                            "disposed unprocessed inode(s) in iput queue when stopping log writer");
2743                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2744         }
2745
2746         WARN_ON(!list_empty(&sci->sc_segbufs));
2747         WARN_ON(!list_empty(&sci->sc_write_logs));
2748
2749         nilfs_put_root(sci->sc_root);
2750
2751         down_write(&nilfs->ns_segctor_sem);
2752
2753         del_timer_sync(&sci->sc_timer);
2754         kfree(sci);
2755 }
2756
2757 /**
2758  * nilfs_attach_log_writer - attach log writer
2759  * @sb: super block instance
2760  * @root: root object of the current filesystem tree
2761  *
2762  * This allocates a log writer object, initializes it, and starts the
2763  * log writer.
2764  *
2765  * Return Value: On success, 0 is returned. On error, one of the following
2766  * negative error code is returned.
2767  *
2768  * %-ENOMEM - Insufficient memory available.
2769  */
2770 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2771 {
2772         struct the_nilfs *nilfs = sb->s_fs_info;
2773         int err;
2774
2775         if (nilfs->ns_writer) {
2776                 /*
2777                  * This happens if the filesystem was remounted
2778                  * read/write after nilfs_error degenerated it into a
2779                  * read-only mount.
2780                  */
2781                 nilfs_detach_log_writer(sb);
2782         }
2783
2784         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2785         if (!nilfs->ns_writer)
2786                 return -ENOMEM;
2787
2788         inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2789
2790         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2791         if (err) {
2792                 kfree(nilfs->ns_writer);
2793                 nilfs->ns_writer = NULL;
2794         }
2795         return err;
2796 }
2797
2798 /**
2799  * nilfs_detach_log_writer - destroy log writer
2800  * @sb: super block instance
2801  *
2802  * This kills log writer daemon, frees the log writer object, and
2803  * destroys list of dirty files.
2804  */
2805 void nilfs_detach_log_writer(struct super_block *sb)
2806 {
2807         struct the_nilfs *nilfs = sb->s_fs_info;
2808         LIST_HEAD(garbage_list);
2809
2810         down_write(&nilfs->ns_segctor_sem);
2811         if (nilfs->ns_writer) {
2812                 nilfs_segctor_destroy(nilfs->ns_writer);
2813                 nilfs->ns_writer = NULL;
2814         }
2815
2816         /* Force to free the list of dirty files */
2817         spin_lock(&nilfs->ns_inode_lock);
2818         if (!list_empty(&nilfs->ns_dirty_files)) {
2819                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2820                 nilfs_warn(sb,
2821                            "disposed unprocessed dirty file(s) when detaching log writer");
2822         }
2823         spin_unlock(&nilfs->ns_inode_lock);
2824         up_write(&nilfs->ns_segctor_sem);
2825
2826         nilfs_dispose_list(nilfs, &garbage_list, 1);
2827 }