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 inode *btnc_inode = ii->i_assoc_inode;
 742         struct pagevec pvec;
 743         struct buffer_head *bh, *head;
 744         unsigned int i;
 745         pgoff_t index = 0;
 746
 747         if (!btnc_inode)
 748                 return;
 749
 750         pagevec_init(&pvec);
 751
 752         while (pagevec_lookup_tag(&pvec, btnc_inode->i_mapping, &index,
 753                                         PAGECACHE_TAG_DIRTY)) {
 754                 for (i = 0; i < pagevec_count(&pvec); i++) {
 755                         bh = head = page_buffers(pvec.pages[i]);
 756                         do {
 757                                 if (buffer_dirty(bh) &&
 758                                                 !buffer_async_write(bh)) {
 759                                         get_bh(bh);
 760                                         list_add_tail(&bh->b_assoc_buffers,
 761                                                       listp);
 762                                 }
 763                                 bh = bh->b_this_page;
 764                         } while (bh != head);
 765                 }
 766                 pagevec_release(&pvec);
 767                 cond_resched();
 768         }
 769 }
 770
 771 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 772                                struct list_head *head, int force)
 773 {
 774         struct nilfs_inode_info *ii, *n;
 775         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 776         unsigned int nv = 0;
 777
 778         while (!list_empty(head)) {
 779                 spin_lock(&nilfs->ns_inode_lock);
 780                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 781                         list_del_init(&ii->i_dirty);
 782                         if (force) {
 783                                 if (unlikely(ii->i_bh)) {
 784                                         brelse(ii->i_bh);
 785                                         ii->i_bh = NULL;
 786                                 }
 787                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 788                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 789                                 list_add_tail(&ii->i_dirty,
 790                                               &nilfs->ns_dirty_files);
 791                                 continue;
 792                         }
 793                         ivec[nv++] = ii;
 794                         if (nv == SC_N_INODEVEC)
 795                                 break;
 796                 }
 797                 spin_unlock(&nilfs->ns_inode_lock);
 798
 799                 for (pii = ivec; nv > 0; pii++, nv--)
 800                         iput(&(*pii)->vfs_inode);
 801         }
 802 }
 803
 804 static void nilfs_iput_work_func(struct work_struct *work)
 805 {
 806         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 807                                                  sc_iput_work);
 808         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 809
 810         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 811 }
 812
 813 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 814                                      struct nilfs_root *root)
 815 {
 816         int ret = 0;
 817
 818         if (nilfs_mdt_fetch_dirty(root->ifile))
 819                 ret++;
 820         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 821                 ret++;
 822         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 823                 ret++;
 824         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 825                 ret++;
 826         return ret;
 827 }
 828
 829 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 830 {
 831         return list_empty(&sci->sc_dirty_files) &&
 832                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 833                 sci->sc_nfreesegs == 0 &&
 834                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 835 }
 836
 837 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 838 {
 839         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 840         int ret = 0;
 841
 842         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 843                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 844
 845         spin_lock(&nilfs->ns_inode_lock);
 846         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 847                 ret++;
 848
 849         spin_unlock(&nilfs->ns_inode_lock);
 850         return ret;
 851 }
 852
 853 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 854 {
 855         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 856
 857         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 858         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 859         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 860         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 861 }
 862
 863 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 864 {
 865         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 866         struct buffer_head *bh_cp;
 867         struct nilfs_checkpoint *raw_cp;
 868         int err;
 869
 870         /* XXX: this interface will be changed */
 871         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 872                                           &raw_cp, &bh_cp);
 873         if (likely(!err)) {
 874                 /*
 875                  * The following code is duplicated with cpfile.  But, it is
 876                  * needed to collect the checkpoint even if it was not newly
 877                  * created.
 878                  */
 879                 mark_buffer_dirty(bh_cp);
 880                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 881                 nilfs_cpfile_put_checkpoint(
 882                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 883         } else if (err == -EINVAL || err == -ENOENT) {
 884                 nilfs_error(sci->sc_super,
 885                             "checkpoint creation failed due to metadata corruption.");
 886                 err = -EIO;
 887         }
 888         return err;
 889 }
 890
 891 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 892 {
 893         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 894         struct buffer_head *bh_cp;
 895         struct nilfs_checkpoint *raw_cp;
 896         int err;
 897
 898         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 899                                           &raw_cp, &bh_cp);
 900         if (unlikely(err)) {
 901                 if (err == -EINVAL || err == -ENOENT) {
 902                         nilfs_error(sci->sc_super,
 903                                     "checkpoint finalization failed due to metadata corruption.");
 904                         err = -EIO;
 905                 }
 906                 goto failed_ibh;
 907         }
 908         raw_cp->cp_snapshot_list.ssl_next = 0;
 909         raw_cp->cp_snapshot_list.ssl_prev = 0;
 910         raw_cp->cp_inodes_count =
 911                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 912         raw_cp->cp_blocks_count =
 913                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 914         raw_cp->cp_nblk_inc =
 915                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 916         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 917         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 918
 919         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 920                 nilfs_checkpoint_clear_minor(raw_cp);
 921         else
 922                 nilfs_checkpoint_set_minor(raw_cp);
 923
 924         nilfs_write_inode_common(sci->sc_root->ifile,
 925                                  &raw_cp->cp_ifile_inode, 1);
 926         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 927         return 0;
 928
 929  failed_ibh:
 930         return err;
 931 }
 932
 933 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 934                                     struct nilfs_inode_info *ii)
 935
 936 {
 937         struct buffer_head *ibh;
 938         struct nilfs_inode *raw_inode;
 939
 940         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 941                 ibh = ii->i_bh;
 942                 BUG_ON(!ibh);
 943                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 944                                                   ibh);
 945                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 946                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 947         }
 948 }
 949
 950 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 951 {
 952         struct nilfs_inode_info *ii;
 953
 954         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 955                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 956                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 957         }
 958 }
 959
 960 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 961                                              struct the_nilfs *nilfs)
 962 {
 963         struct buffer_head *bh_sr;
 964         struct nilfs_super_root *raw_sr;
 965         unsigned int isz, srsz;
 966
 967         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 968         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 969         isz = nilfs->ns_inode_size;
 970         srsz = NILFS_SR_BYTES(isz);
 971
 972         raw_sr->sr_bytes = cpu_to_le16(srsz);
 973         raw_sr->sr_nongc_ctime
 974                 = cpu_to_le64(nilfs_doing_gc() ?
 975                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 976         raw_sr->sr_flags = 0;
 977
 978         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 979                                  NILFS_SR_DAT_OFFSET(isz), 1);
 980         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 981                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 982         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 983                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 984         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 985 }
 986
 987 static void nilfs_redirty_inodes(struct list_head *head)
 988 {
 989         struct nilfs_inode_info *ii;
 990
 991         list_for_each_entry(ii, head, i_dirty) {
 992                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 993                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 994         }
 995 }
 996
 997 static void nilfs_drop_collected_inodes(struct list_head *head)
 998 {
 999         struct nilfs_inode_info *ii;
1000
1001         list_for_each_entry(ii, head, i_dirty) {
1002                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1003                         continue;
1004
1005                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1006                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1007         }
1008 }
1009
1010 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1011                                        struct inode *inode,
1012                                        struct list_head *listp,
1013                                        int (*collect)(struct nilfs_sc_info *,
1014                                                       struct buffer_head *,
1015                                                       struct inode *))
1016 {
1017         struct buffer_head *bh, *n;
1018         int err = 0;
1019
1020         if (collect) {
1021                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1022                         list_del_init(&bh->b_assoc_buffers);
1023                         err = collect(sci, bh, inode);
1024                         brelse(bh);
1025                         if (unlikely(err))
1026                                 goto dispose_buffers;
1027                 }
1028                 return 0;
1029         }
1030
1031  dispose_buffers:
1032         while (!list_empty(listp)) {
1033                 bh = list_first_entry(listp, struct buffer_head,
1034                                       b_assoc_buffers);
1035                 list_del_init(&bh->b_assoc_buffers);
1036                 brelse(bh);
1037         }
1038         return err;
1039 }
1040
1041 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1042 {
1043         /* Remaining number of blocks within segment buffer */
1044         return sci->sc_segbuf_nblocks -
1045                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1046 }
1047
1048 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1049                                    struct inode *inode,
1050                                    const struct nilfs_sc_operations *sc_ops)
1051 {
1052         LIST_HEAD(data_buffers);
1053         LIST_HEAD(node_buffers);
1054         int err;
1055
1056         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1057                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1058
1059                 n = nilfs_lookup_dirty_data_buffers(
1060                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1061                 if (n > rest) {
1062                         err = nilfs_segctor_apply_buffers(
1063                                 sci, inode, &data_buffers,
1064                                 sc_ops->collect_data);
1065                         BUG_ON(!err); /* always receive -E2BIG or true error */
1066                         goto break_or_fail;
1067                 }
1068         }
1069         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1070
1071         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1072                 err = nilfs_segctor_apply_buffers(
1073                         sci, inode, &data_buffers, sc_ops->collect_data);
1074                 if (unlikely(err)) {
1075                         /* dispose node list */
1076                         nilfs_segctor_apply_buffers(
1077                                 sci, inode, &node_buffers, NULL);
1078                         goto break_or_fail;
1079                 }
1080                 sci->sc_stage.flags |= NILFS_CF_NODE;
1081         }
1082         /* Collect node */
1083         err = nilfs_segctor_apply_buffers(
1084                 sci, inode, &node_buffers, sc_ops->collect_node);
1085         if (unlikely(err))
1086                 goto break_or_fail;
1087
1088         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1089         err = nilfs_segctor_apply_buffers(
1090                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1091         if (unlikely(err))
1092                 goto break_or_fail;
1093
1094         nilfs_segctor_end_finfo(sci, inode);
1095         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1096
1097  break_or_fail:
1098         return err;
1099 }
1100
1101 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1102                                          struct inode *inode)
1103 {
1104         LIST_HEAD(data_buffers);
1105         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1106         int err;
1107
1108         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1109                                             sci->sc_dsync_start,
1110                                             sci->sc_dsync_end);
1111
1112         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1113                                           nilfs_collect_file_data);
1114         if (!err) {
1115                 nilfs_segctor_end_finfo(sci, inode);
1116                 BUG_ON(n > rest);
1117                 /* always receive -E2BIG or true error if n > rest */
1118         }
1119         return err;
1120 }
1121
1122 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1123 {
1124         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1125         struct list_head *head;
1126         struct nilfs_inode_info *ii;
1127         size_t ndone;
1128         int err = 0;
1129
1130         switch (nilfs_sc_cstage_get(sci)) {
1131         case NILFS_ST_INIT:
1132                 /* Pre-processes */
1133                 sci->sc_stage.flags = 0;
1134
1135                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1136                         sci->sc_nblk_inc = 0;
1137                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1138                         if (mode == SC_LSEG_DSYNC) {
1139                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1140                                 goto dsync_mode;
1141                         }
1142                 }
1143
1144                 sci->sc_stage.dirty_file_ptr = NULL;
1145                 sci->sc_stage.gc_inode_ptr = NULL;
1146                 if (mode == SC_FLUSH_DAT) {
1147                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1148                         goto dat_stage;
1149                 }
1150                 nilfs_sc_cstage_inc(sci);
1151                 fallthrough;
1152         case NILFS_ST_GC:
1153                 if (nilfs_doing_gc()) {
1154                         head = &sci->sc_gc_inodes;
1155                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1156                                                 head, i_dirty);
1157                         list_for_each_entry_continue(ii, head, i_dirty) {
1158                                 err = nilfs_segctor_scan_file(
1159                                         sci, &ii->vfs_inode,
1160                                         &nilfs_sc_file_ops);
1161                                 if (unlikely(err)) {
1162                                         sci->sc_stage.gc_inode_ptr = list_entry(
1163                                                 ii->i_dirty.prev,
1164                                                 struct nilfs_inode_info,
1165                                                 i_dirty);
1166                                         goto break_or_fail;
1167                                 }
1168                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1169                         }
1170                         sci->sc_stage.gc_inode_ptr = NULL;
1171                 }
1172                 nilfs_sc_cstage_inc(sci);
1173                 fallthrough;
1174         case NILFS_ST_FILE:
1175                 head = &sci->sc_dirty_files;
1176                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1177                                         i_dirty);
1178                 list_for_each_entry_continue(ii, head, i_dirty) {
1179                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1180
1181                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1182                                                       &nilfs_sc_file_ops);
1183                         if (unlikely(err)) {
1184                                 sci->sc_stage.dirty_file_ptr =
1185                                         list_entry(ii->i_dirty.prev,
1186                                                    struct nilfs_inode_info,
1187                                                    i_dirty);
1188                                 goto break_or_fail;
1189                         }
1190                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1191                         /* XXX: required ? */
1192                 }
1193                 sci->sc_stage.dirty_file_ptr = NULL;
1194                 if (mode == SC_FLUSH_FILE) {
1195                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1196                         return 0;
1197                 }
1198                 nilfs_sc_cstage_inc(sci);
1199                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1200                 fallthrough;
1201         case NILFS_ST_IFILE:
1202                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1203                                               &nilfs_sc_file_ops);
1204                 if (unlikely(err))
1205                         break;
1206                 nilfs_sc_cstage_inc(sci);
1207                 /* Creating a checkpoint */
1208                 err = nilfs_segctor_create_checkpoint(sci);
1209                 if (unlikely(err))
1210                         break;
1211                 fallthrough;
1212         case NILFS_ST_CPFILE:
1213                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1214                                               &nilfs_sc_file_ops);
1215                 if (unlikely(err))
1216                         break;
1217                 nilfs_sc_cstage_inc(sci);
1218                 fallthrough;
1219         case NILFS_ST_SUFILE:
1220                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1221                                          sci->sc_nfreesegs, &ndone);
1222                 if (unlikely(err)) {
1223                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1224                                                   sci->sc_freesegs, ndone,
1225                                                   NULL);
1226                         break;
1227                 }
1228                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1229
1230                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1231                                               &nilfs_sc_file_ops);
1232                 if (unlikely(err))
1233                         break;
1234                 nilfs_sc_cstage_inc(sci);
1235                 fallthrough;
1236         case NILFS_ST_DAT:
1237  dat_stage:
1238                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1239                                               &nilfs_sc_dat_ops);
1240                 if (unlikely(err))
1241                         break;
1242                 if (mode == SC_FLUSH_DAT) {
1243                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1244                         return 0;
1245                 }
1246                 nilfs_sc_cstage_inc(sci);
1247                 fallthrough;
1248         case NILFS_ST_SR:
1249                 if (mode == SC_LSEG_SR) {
1250                         /* Appending a super root */
1251                         err = nilfs_segctor_add_super_root(sci);
1252                         if (unlikely(err))
1253                                 break;
1254                 }
1255                 /* End of a logical segment */
1256                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1257                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1258                 return 0;
1259         case NILFS_ST_DSYNC:
1260  dsync_mode:
1261                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1262                 ii = sci->sc_dsync_inode;
1263                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1264                         break;
1265
1266                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1267                 if (unlikely(err))
1268                         break;
1269                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1270                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1271                 return 0;
1272         case NILFS_ST_DONE:
1273                 return 0;
1274         default:
1275                 BUG();
1276         }
1277
1278  break_or_fail:
1279         return err;
1280 }
1281
1282 /**
1283  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1284  * @sci: nilfs_sc_info
1285  * @nilfs: nilfs object
1286  */
1287 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1288                                             struct the_nilfs *nilfs)
1289 {
1290         struct nilfs_segment_buffer *segbuf, *prev;
1291         __u64 nextnum;
1292         int err, alloc = 0;
1293
1294         segbuf = nilfs_segbuf_new(sci->sc_super);
1295         if (unlikely(!segbuf))
1296                 return -ENOMEM;
1297
1298         if (list_empty(&sci->sc_write_logs)) {
1299                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1300                                  nilfs->ns_pseg_offset, nilfs);
1301                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1302                         nilfs_shift_to_next_segment(nilfs);
1303                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1304                 }
1305
1306                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1307                 nextnum = nilfs->ns_nextnum;
1308
1309                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1310                         /* Start from the head of a new full segment */
1311                         alloc++;
1312         } else {
1313                 /* Continue logs */
1314                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1315                 nilfs_segbuf_map_cont(segbuf, prev);
1316                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1317                 nextnum = prev->sb_nextnum;
1318
1319                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1320                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1321                         segbuf->sb_sum.seg_seq++;
1322                         alloc++;
1323                 }
1324         }
1325
1326         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1327         if (err)
1328                 goto failed;
1329
1330         if (alloc) {
1331                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1332                 if (err)
1333                         goto failed;
1334         }
1335         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1336
1337         BUG_ON(!list_empty(&sci->sc_segbufs));
1338         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1339         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1340         return 0;
1341
1342  failed:
1343         nilfs_segbuf_free(segbuf);
1344         return err;
1345 }
1346
1347 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1348                                          struct the_nilfs *nilfs, int nadd)
1349 {
1350         struct nilfs_segment_buffer *segbuf, *prev;
1351         struct inode *sufile = nilfs->ns_sufile;
1352         __u64 nextnextnum;
1353         LIST_HEAD(list);
1354         int err, ret, i;
1355
1356         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1357         /*
1358          * Since the segment specified with nextnum might be allocated during
1359          * the previous construction, the buffer including its segusage may
1360          * not be dirty.  The following call ensures that the buffer is dirty
1361          * and will pin the buffer on memory until the sufile is written.
1362          */
1363         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1364         if (unlikely(err))
1365                 return err;
1366
1367         for (i = 0; i < nadd; i++) {
1368                 /* extend segment info */
1369                 err = -ENOMEM;
1370                 segbuf = nilfs_segbuf_new(sci->sc_super);
1371                 if (unlikely(!segbuf))
1372                         goto failed;
1373
1374                 /* map this buffer to region of segment on-disk */
1375                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1376                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1377
1378                 /* allocate the next next full segment */
1379                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1380                 if (unlikely(err))
1381                         goto failed_segbuf;
1382
1383                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1384                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1385
1386                 list_add_tail(&segbuf->sb_list, &list);
1387                 prev = segbuf;
1388         }
1389         list_splice_tail(&list, &sci->sc_segbufs);
1390         return 0;
1391
1392  failed_segbuf:
1393         nilfs_segbuf_free(segbuf);
1394  failed:
1395         list_for_each_entry(segbuf, &list, sb_list) {
1396                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1397                 WARN_ON(ret); /* never fails */
1398         }
1399         nilfs_destroy_logs(&list);
1400         return err;
1401 }
1402
1403 static void nilfs_free_incomplete_logs(struct list_head *logs,
1404                                        struct the_nilfs *nilfs)
1405 {
1406         struct nilfs_segment_buffer *segbuf, *prev;
1407         struct inode *sufile = nilfs->ns_sufile;
1408         int ret;
1409
1410         segbuf = NILFS_FIRST_SEGBUF(logs);
1411         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1412                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1413                 WARN_ON(ret); /* never fails */
1414         }
1415         if (atomic_read(&segbuf->sb_err)) {
1416                 /* Case 1: The first segment failed */
1417                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1418                         /*
1419                          * Case 1a:  Partial segment appended into an existing
1420                          * segment
1421                          */
1422                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1423                                                 segbuf->sb_fseg_end);
1424                 else /* Case 1b:  New full segment */
1425                         set_nilfs_discontinued(nilfs);
1426         }
1427
1428         prev = segbuf;
1429         list_for_each_entry_continue(segbuf, logs, sb_list) {
1430                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1431                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1432                         WARN_ON(ret); /* never fails */
1433                 }
1434                 if (atomic_read(&segbuf->sb_err) &&
1435                     segbuf->sb_segnum != nilfs->ns_nextnum)
1436                         /* Case 2: extended segment (!= next) failed */
1437                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1438                 prev = segbuf;
1439         }
1440 }
1441
1442 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1443                                           struct inode *sufile)
1444 {
1445         struct nilfs_segment_buffer *segbuf;
1446         unsigned long live_blocks;
1447         int ret;
1448
1449         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1450                 live_blocks = segbuf->sb_sum.nblocks +
1451                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1452                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1453                                                      live_blocks,
1454                                                      sci->sc_seg_ctime);
1455                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1456         }
1457 }
1458
1459 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1460 {
1461         struct nilfs_segment_buffer *segbuf;
1462         int ret;
1463
1464         segbuf = NILFS_FIRST_SEGBUF(logs);
1465         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1466                                              segbuf->sb_pseg_start -
1467                                              segbuf->sb_fseg_start, 0);
1468         WARN_ON(ret); /* always succeed because the segusage is dirty */
1469
1470         list_for_each_entry_continue(segbuf, logs, sb_list) {
1471                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1472                                                      0, 0);
1473                 WARN_ON(ret); /* always succeed */
1474         }
1475 }
1476
1477 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1478                                             struct nilfs_segment_buffer *last,
1479                                             struct inode *sufile)
1480 {
1481         struct nilfs_segment_buffer *segbuf = last;
1482         int ret;
1483
1484         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1485                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1486                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1487                 WARN_ON(ret);
1488         }
1489         nilfs_truncate_logs(&sci->sc_segbufs, last);
1490 }
1491
1492
1493 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1494                                  struct the_nilfs *nilfs, int mode)
1495 {
1496         struct nilfs_cstage prev_stage = sci->sc_stage;
1497         int err, nadd = 1;
1498
1499         /* Collection retry loop */
1500         for (;;) {
1501                 sci->sc_nblk_this_inc = 0;
1502                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1503
1504                 err = nilfs_segctor_reset_segment_buffer(sci);
1505                 if (unlikely(err))
1506                         goto failed;
1507
1508                 err = nilfs_segctor_collect_blocks(sci, mode);
1509                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1510                 if (!err)
1511                         break;
1512
1513                 if (unlikely(err != -E2BIG))
1514                         goto failed;
1515
1516                 /* The current segment is filled up */
1517                 if (mode != SC_LSEG_SR ||
1518                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1519                         break;
1520
1521                 nilfs_clear_logs(&sci->sc_segbufs);
1522
1523                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1524                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1525                                                         sci->sc_freesegs,
1526                                                         sci->sc_nfreesegs,
1527                                                         NULL);
1528                         WARN_ON(err); /* do not happen */
1529                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1530                 }
1531
1532                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1533                 if (unlikely(err))
1534                         return err;
1535
1536                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1537                 sci->sc_stage = prev_stage;
1538         }
1539         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1540         return 0;
1541
1542  failed:
1543         return err;
1544 }
1545
1546 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1547                                       struct buffer_head *new_bh)
1548 {
1549         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1550
1551         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1552         /* The caller must release old_bh */
1553 }
1554
1555 static int
1556 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1557                                      struct nilfs_segment_buffer *segbuf,
1558                                      int mode)
1559 {
1560         struct inode *inode = NULL;
1561         sector_t blocknr;
1562         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1563         unsigned long nblocks = 0, ndatablk = 0;
1564         const struct nilfs_sc_operations *sc_op = NULL;
1565         struct nilfs_segsum_pointer ssp;
1566         struct nilfs_finfo *finfo = NULL;
1567         union nilfs_binfo binfo;
1568         struct buffer_head *bh, *bh_org;
1569         ino_t ino = 0;
1570         int err = 0;
1571
1572         if (!nfinfo)
1573                 goto out;
1574
1575         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1576         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1577         ssp.offset = sizeof(struct nilfs_segment_summary);
1578
1579         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1580                 if (bh == segbuf->sb_super_root)
1581                         break;
1582                 if (!finfo) {
1583                         finfo = nilfs_segctor_map_segsum_entry(
1584                                 sci, &ssp, sizeof(*finfo));
1585                         ino = le64_to_cpu(finfo->fi_ino);
1586                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1587                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1588
1589                         inode = bh->b_page->mapping->host;
1590
1591                         if (mode == SC_LSEG_DSYNC)
1592                                 sc_op = &nilfs_sc_dsync_ops;
1593                         else if (ino == NILFS_DAT_INO)
1594                                 sc_op = &nilfs_sc_dat_ops;
1595                         else /* file blocks */
1596                                 sc_op = &nilfs_sc_file_ops;
1597                 }
1598                 bh_org = bh;
1599                 get_bh(bh_org);
1600                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1601                                         &binfo);
1602                 if (bh != bh_org)
1603                         nilfs_list_replace_buffer(bh_org, bh);
1604                 brelse(bh_org);
1605                 if (unlikely(err))
1606                         goto failed_bmap;
1607
1608                 if (ndatablk > 0)
1609                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1610                 else
1611                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1612
1613                 blocknr++;
1614                 if (--nblocks == 0) {
1615                         finfo = NULL;
1616                         if (--nfinfo == 0)
1617                                 break;
1618                 } else if (ndatablk > 0)
1619                         ndatablk--;
1620         }
1621  out:
1622         return 0;
1623
1624  failed_bmap:
1625         return err;
1626 }
1627
1628 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1629 {
1630         struct nilfs_segment_buffer *segbuf;
1631         int err;
1632
1633         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1634                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1635                 if (unlikely(err))
1636                         return err;
1637                 nilfs_segbuf_fill_in_segsum(segbuf);
1638         }
1639         return 0;
1640 }
1641
1642 static void nilfs_begin_page_io(struct page *page)
1643 {
1644         if (!page || PageWriteback(page))
1645                 /*
1646                  * For split b-tree node pages, this function may be called
1647                  * twice.  We ignore the 2nd or later calls by this check.
1648                  */
1649                 return;
1650
1651         lock_page(page);
1652         clear_page_dirty_for_io(page);
1653         set_page_writeback(page);
1654         unlock_page(page);
1655 }
1656
1657 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1658 {
1659         struct nilfs_segment_buffer *segbuf;
1660         struct page *bd_page = NULL, *fs_page = NULL;
1661
1662         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1663                 struct buffer_head *bh;
1664
1665                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1666                                     b_assoc_buffers) {
1667                         if (bh->b_page != bd_page) {
1668                                 if (bd_page) {
1669                                         lock_page(bd_page);
1670                                         clear_page_dirty_for_io(bd_page);
1671                                         set_page_writeback(bd_page);
1672                                         unlock_page(bd_page);
1673                                 }
1674                                 bd_page = bh->b_page;
1675                         }
1676                 }
1677
1678                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1679                                     b_assoc_buffers) {
1680                         set_buffer_async_write(bh);
1681                         if (bh == segbuf->sb_super_root) {
1682                                 if (bh->b_page != bd_page) {
1683                                         lock_page(bd_page);
1684                                         clear_page_dirty_for_io(bd_page);
1685                                         set_page_writeback(bd_page);
1686                                         unlock_page(bd_page);
1687                                         bd_page = bh->b_page;
1688                                 }
1689                                 break;
1690                         }
1691                         if (bh->b_page != fs_page) {
1692                                 nilfs_begin_page_io(fs_page);
1693                                 fs_page = bh->b_page;
1694                         }
1695                 }
1696         }
1697         if (bd_page) {
1698                 lock_page(bd_page);
1699                 clear_page_dirty_for_io(bd_page);
1700                 set_page_writeback(bd_page);
1701                 unlock_page(bd_page);
1702         }
1703         nilfs_begin_page_io(fs_page);
1704 }
1705
1706 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1707                                struct the_nilfs *nilfs)
1708 {
1709         int ret;
1710
1711         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1712         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1713         return ret;
1714 }
1715
1716 static void nilfs_end_page_io(struct page *page, int err)
1717 {
1718         if (!page)
1719                 return;
1720
1721         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1722                 /*
1723                  * For b-tree node pages, this function may be called twice
1724                  * or more because they might be split in a segment.
1725                  */
1726                 if (PageDirty(page)) {
1727                         /*
1728                          * For pages holding split b-tree node buffers, dirty
1729                          * flag on the buffers may be cleared discretely.
1730                          * In that case, the page is once redirtied for
1731                          * remaining buffers, and it must be cancelled if
1732                          * all the buffers get cleaned later.
1733                          */
1734                         lock_page(page);
1735                         if (nilfs_page_buffers_clean(page))
1736                                 __nilfs_clear_page_dirty(page);
1737                         unlock_page(page);
1738                 }
1739                 return;
1740         }
1741
1742         if (!err) {
1743                 if (!nilfs_page_buffers_clean(page))
1744                         __set_page_dirty_nobuffers(page);
1745                 ClearPageError(page);
1746         } else {
1747                 __set_page_dirty_nobuffers(page);
1748                 SetPageError(page);
1749         }
1750
1751         end_page_writeback(page);
1752 }
1753
1754 static void nilfs_abort_logs(struct list_head *logs, int err)
1755 {
1756         struct nilfs_segment_buffer *segbuf;
1757         struct page *bd_page = NULL, *fs_page = NULL;
1758         struct buffer_head *bh;
1759
1760         if (list_empty(logs))
1761                 return;
1762
1763         list_for_each_entry(segbuf, logs, sb_list) {
1764                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1765                                     b_assoc_buffers) {
1766                         if (bh->b_page != bd_page) {
1767                                 if (bd_page)
1768                                         end_page_writeback(bd_page);
1769                                 bd_page = bh->b_page;
1770                         }
1771                 }
1772
1773                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1774                                     b_assoc_buffers) {
1775                         clear_buffer_async_write(bh);
1776                         if (bh == segbuf->sb_super_root) {
1777                                 if (bh->b_page != bd_page) {
1778                                         end_page_writeback(bd_page);
1779                                         bd_page = bh->b_page;
1780                                 }
1781                                 break;
1782                         }
1783                         if (bh->b_page != fs_page) {
1784                                 nilfs_end_page_io(fs_page, err);
1785                                 fs_page = bh->b_page;
1786                         }
1787                 }
1788         }
1789         if (bd_page)
1790                 end_page_writeback(bd_page);
1791
1792         nilfs_end_page_io(fs_page, err);
1793 }
1794
1795 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1796                                              struct the_nilfs *nilfs, int err)
1797 {
1798         LIST_HEAD(logs);
1799         int ret;
1800
1801         list_splice_tail_init(&sci->sc_write_logs, &logs);
1802         ret = nilfs_wait_on_logs(&logs);
1803         nilfs_abort_logs(&logs, ret ? : err);
1804
1805         list_splice_tail_init(&sci->sc_segbufs, &logs);
1806         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1807         nilfs_free_incomplete_logs(&logs, nilfs);
1808
1809         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1810                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1811                                                 sci->sc_freesegs,
1812                                                 sci->sc_nfreesegs,
1813                                                 NULL);
1814                 WARN_ON(ret); /* do not happen */
1815         }
1816
1817         nilfs_destroy_logs(&logs);
1818 }
1819
1820 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1821                                    struct nilfs_segment_buffer *segbuf)
1822 {
1823         nilfs->ns_segnum = segbuf->sb_segnum;
1824         nilfs->ns_nextnum = segbuf->sb_nextnum;
1825         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1826                 + segbuf->sb_sum.nblocks;
1827         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1828         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1829 }
1830
1831 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1832 {
1833         struct nilfs_segment_buffer *segbuf;
1834         struct page *bd_page = NULL, *fs_page = NULL;
1835         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1836         int update_sr = false;
1837
1838         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1839                 struct buffer_head *bh;
1840
1841                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1842                                     b_assoc_buffers) {
1843                         set_buffer_uptodate(bh);
1844                         clear_buffer_dirty(bh);
1845                         if (bh->b_page != bd_page) {
1846                                 if (bd_page)
1847                                         end_page_writeback(bd_page);
1848                                 bd_page = bh->b_page;
1849                         }
1850                 }
1851                 /*
1852                  * We assume that the buffers which belong to the same page
1853                  * continue over the buffer list.
1854                  * Under this assumption, the last BHs of pages is
1855                  * identifiable by the discontinuity of bh->b_page
1856                  * (page != fs_page).
1857                  *
1858                  * For B-tree node blocks, however, this assumption is not
1859                  * guaranteed.  The cleanup code of B-tree node pages needs
1860                  * special care.
1861                  */
1862                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1863                                     b_assoc_buffers) {
1864                         const unsigned long set_bits = BIT(BH_Uptodate);
1865                         const unsigned long clear_bits =
1866                                 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1867                                  BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1868                                  BIT(BH_NILFS_Redirected));
1869
1870                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1871                         if (bh == segbuf->sb_super_root) {
1872                                 if (bh->b_page != bd_page) {
1873                                         end_page_writeback(bd_page);
1874                                         bd_page = bh->b_page;
1875                                 }
1876                                 update_sr = true;
1877                                 break;
1878                         }
1879                         if (bh->b_page != fs_page) {
1880                                 nilfs_end_page_io(fs_page, 0);
1881                                 fs_page = bh->b_page;
1882                         }
1883                 }
1884
1885                 if (!nilfs_segbuf_simplex(segbuf)) {
1886                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1887                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1888                                 sci->sc_lseg_stime = jiffies;
1889                         }
1890                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1891                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1892                 }
1893         }
1894         /*
1895          * Since pages may continue over multiple segment buffers,
1896          * end of the last page must be checked outside of the loop.
1897          */
1898         if (bd_page)
1899                 end_page_writeback(bd_page);
1900
1901         nilfs_end_page_io(fs_page, 0);
1902
1903         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1904
1905         if (nilfs_doing_gc())
1906                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1907         else
1908                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1909
1910         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1911
1912         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1913         nilfs_set_next_segment(nilfs, segbuf);
1914
1915         if (update_sr) {
1916                 nilfs->ns_flushed_device = 0;
1917                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1918                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1919
1920                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1921                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1922                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1923                 nilfs_segctor_clear_metadata_dirty(sci);
1924         } else
1925                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1926 }
1927
1928 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1929 {
1930         int ret;
1931
1932         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1933         if (!ret) {
1934                 nilfs_segctor_complete_write(sci);
1935                 nilfs_destroy_logs(&sci->sc_write_logs);
1936         }
1937         return ret;
1938 }
1939
1940 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1941                                              struct the_nilfs *nilfs)
1942 {
1943         struct nilfs_inode_info *ii, *n;
1944         struct inode *ifile = sci->sc_root->ifile;
1945
1946         spin_lock(&nilfs->ns_inode_lock);
1947  retry:
1948         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1949                 if (!ii->i_bh) {
1950                         struct buffer_head *ibh;
1951                         int err;
1952
1953                         spin_unlock(&nilfs->ns_inode_lock);
1954                         err = nilfs_ifile_get_inode_block(
1955                                 ifile, ii->vfs_inode.i_ino, &ibh);
1956                         if (unlikely(err)) {
1957                                 nilfs_warn(sci->sc_super,
1958                                            "log writer: error %d getting inode block (ino=%lu)",
1959                                            err, ii->vfs_inode.i_ino);
1960                                 return err;
1961                         }
1962                         spin_lock(&nilfs->ns_inode_lock);
1963                         if (likely(!ii->i_bh))
1964                                 ii->i_bh = ibh;
1965                         else
1966                                 brelse(ibh);
1967                         goto retry;
1968                 }
1969
1970                 // Always redirty the buffer to avoid race condition
1971                 mark_buffer_dirty(ii->i_bh);
1972                 nilfs_mdt_mark_dirty(ifile);
1973
1974                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1975                 set_bit(NILFS_I_BUSY, &ii->i_state);
1976                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1977         }
1978         spin_unlock(&nilfs->ns_inode_lock);
1979
1980         return 0;
1981 }
1982
1983 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1984                                              struct the_nilfs *nilfs)
1985 {
1986         struct nilfs_inode_info *ii, *n;
1987         int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
1988         int defer_iput = false;
1989
1990         spin_lock(&nilfs->ns_inode_lock);
1991         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1992                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1993                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1994                         continue;
1995
1996                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1997                 brelse(ii->i_bh);
1998                 ii->i_bh = NULL;
1999                 list_del_init(&ii->i_dirty);
2000                 if (!ii->vfs_inode.i_nlink || during_mount) {
2001                         /*
2002                          * Defer calling iput() to avoid deadlocks if
2003                          * i_nlink == 0 or mount is not yet finished.
2004                          */
2005                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2006                         defer_iput = true;
2007                 } else {
2008                         spin_unlock(&nilfs->ns_inode_lock);
2009                         iput(&ii->vfs_inode);
2010                         spin_lock(&nilfs->ns_inode_lock);
2011                 }
2012         }
2013         spin_unlock(&nilfs->ns_inode_lock);
2014
2015         if (defer_iput)
2016                 schedule_work(&sci->sc_iput_work);
2017 }
2018
2019 /*
2020  * Main procedure of segment constructor
2021  */
2022 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2023 {
2024         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2025         int err;
2026
2027         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2028         sci->sc_cno = nilfs->ns_cno;
2029
2030         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2031         if (unlikely(err))
2032                 goto out;
2033
2034         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2035                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2036
2037         if (nilfs_segctor_clean(sci))
2038                 goto out;
2039
2040         do {
2041                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2042
2043                 err = nilfs_segctor_begin_construction(sci, nilfs);
2044                 if (unlikely(err))
2045                         goto out;
2046
2047                 /* Update time stamp */
2048                 sci->sc_seg_ctime = ktime_get_real_seconds();
2049
2050                 err = nilfs_segctor_collect(sci, nilfs, mode);
2051                 if (unlikely(err))
2052                         goto failed;
2053
2054                 /* Avoid empty segment */
2055                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2056                     nilfs_segbuf_empty(sci->sc_curseg)) {
2057                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2058                         goto out;
2059                 }
2060
2061                 err = nilfs_segctor_assign(sci, mode);
2062                 if (unlikely(err))
2063                         goto failed;
2064
2065                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2066                         nilfs_segctor_fill_in_file_bmap(sci);
2067
2068                 if (mode == SC_LSEG_SR &&
2069                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2070                         err = nilfs_segctor_fill_in_checkpoint(sci);
2071                         if (unlikely(err))
2072                                 goto failed_to_write;
2073
2074                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2075                 }
2076                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2077
2078                 /* Write partial segments */
2079                 nilfs_segctor_prepare_write(sci);
2080
2081                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2082                                             nilfs->ns_crc_seed);
2083
2084                 err = nilfs_segctor_write(sci, nilfs);
2085                 if (unlikely(err))
2086                         goto failed_to_write;
2087
2088                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2089                     nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2090                         /*
2091                          * At this point, we avoid double buffering
2092                          * for blocksize < pagesize because page dirty
2093                          * flag is turned off during write and dirty
2094                          * buffers are not properly collected for
2095                          * pages crossing over segments.
2096                          */
2097                         err = nilfs_segctor_wait(sci);
2098                         if (err)
2099                                 goto failed_to_write;
2100                 }
2101         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2102
2103  out:
2104         nilfs_segctor_drop_written_files(sci, nilfs);
2105         return err;
2106
2107  failed_to_write:
2108         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2109                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2110
2111  failed:
2112         if (nilfs_doing_gc())
2113                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2114         nilfs_segctor_abort_construction(sci, nilfs, err);
2115         goto out;
2116 }
2117
2118 /**
2119  * nilfs_segctor_start_timer - set timer of background write
2120  * @sci: nilfs_sc_info
2121  *
2122  * If the timer has already been set, it ignores the new request.
2123  * This function MUST be called within a section locking the segment
2124  * semaphore.
2125  */
2126 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2127 {
2128         spin_lock(&sci->sc_state_lock);
2129         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2130                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2131                 add_timer(&sci->sc_timer);
2132                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2133         }
2134         spin_unlock(&sci->sc_state_lock);
2135 }
2136
2137 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2138 {
2139         spin_lock(&sci->sc_state_lock);
2140         if (!(sci->sc_flush_request & BIT(bn))) {
2141                 unsigned long prev_req = sci->sc_flush_request;
2142
2143                 sci->sc_flush_request |= BIT(bn);
2144                 if (!prev_req)
2145                         wake_up(&sci->sc_wait_daemon);
2146         }
2147         spin_unlock(&sci->sc_state_lock);
2148 }
2149
2150 /**
2151  * nilfs_flush_segment - trigger a segment construction for resource control
2152  * @sb: super block
2153  * @ino: inode number of the file to be flushed out.
2154  */
2155 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2156 {
2157         struct the_nilfs *nilfs = sb->s_fs_info;
2158         struct nilfs_sc_info *sci = nilfs->ns_writer;
2159
2160         if (!sci || nilfs_doing_construction())
2161                 return;
2162         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2163                                         /* assign bit 0 to data files */
2164 }
2165
2166 struct nilfs_segctor_wait_request {
2167         wait_queue_entry_t      wq;
2168         __u32           seq;
2169         int             err;
2170         atomic_t        done;
2171 };
2172
2173 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2174 {
2175         struct nilfs_segctor_wait_request wait_req;
2176         int err = 0;
2177
2178         spin_lock(&sci->sc_state_lock);
2179         init_wait(&wait_req.wq);
2180         wait_req.err = 0;
2181         atomic_set(&wait_req.done, 0);
2182         wait_req.seq = ++sci->sc_seq_request;
2183         spin_unlock(&sci->sc_state_lock);
2184
2185         init_waitqueue_entry(&wait_req.wq, current);
2186         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2187         set_current_state(TASK_INTERRUPTIBLE);
2188         wake_up(&sci->sc_wait_daemon);
2189
2190         for (;;) {
2191                 if (atomic_read(&wait_req.done)) {
2192                         err = wait_req.err;
2193                         break;
2194                 }
2195                 if (!signal_pending(current)) {
2196                         schedule();
2197                         continue;
2198                 }
2199                 err = -ERESTARTSYS;
2200                 break;
2201         }
2202         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2203         return err;
2204 }
2205
2206 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2207 {
2208         struct nilfs_segctor_wait_request *wrq, *n;
2209         unsigned long flags;
2210
2211         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2212         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2213                 if (!atomic_read(&wrq->done) &&
2214                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2215                         wrq->err = err;
2216                         atomic_set(&wrq->done, 1);
2217                 }
2218                 if (atomic_read(&wrq->done)) {
2219                         wrq->wq.func(&wrq->wq,
2220                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2221                                      0, NULL);
2222                 }
2223         }
2224         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2225 }
2226
2227 /**
2228  * nilfs_construct_segment - construct a logical segment
2229  * @sb: super block
2230  *
2231  * Return Value: On success, 0 is retured. On errors, one of the following
2232  * negative error code is returned.
2233  *
2234  * %-EROFS - Read only filesystem.
2235  *
2236  * %-EIO - I/O error
2237  *
2238  * %-ENOSPC - No space left on device (only in a panic state).
2239  *
2240  * %-ERESTARTSYS - Interrupted.
2241  *
2242  * %-ENOMEM - Insufficient memory available.
2243  */
2244 int nilfs_construct_segment(struct super_block *sb)
2245 {
2246         struct the_nilfs *nilfs = sb->s_fs_info;
2247         struct nilfs_sc_info *sci = nilfs->ns_writer;
2248         struct nilfs_transaction_info *ti;
2249         int err;
2250
2251         if (!sci)
2252                 return -EROFS;
2253
2254         /* A call inside transactions causes a deadlock. */
2255         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2256
2257         err = nilfs_segctor_sync(sci);
2258         return err;
2259 }
2260
2261 /**
2262  * nilfs_construct_dsync_segment - construct a data-only logical segment
2263  * @sb: super block
2264  * @inode: inode whose data blocks should be written out
2265  * @start: start byte offset
2266  * @end: end byte offset (inclusive)
2267  *
2268  * Return Value: On success, 0 is retured. On errors, one of the following
2269  * negative error code is returned.
2270  *
2271  * %-EROFS - Read only filesystem.
2272  *
2273  * %-EIO - I/O error
2274  *
2275  * %-ENOSPC - No space left on device (only in a panic state).
2276  *
2277  * %-ERESTARTSYS - Interrupted.
2278  *
2279  * %-ENOMEM - Insufficient memory available.
2280  */
2281 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2282                                   loff_t start, loff_t end)
2283 {
2284         struct the_nilfs *nilfs = sb->s_fs_info;
2285         struct nilfs_sc_info *sci = nilfs->ns_writer;
2286         struct nilfs_inode_info *ii;
2287         struct nilfs_transaction_info ti;
2288         int err = 0;
2289
2290         if (!sci)
2291                 return -EROFS;
2292
2293         nilfs_transaction_lock(sb, &ti, 0);
2294
2295         ii = NILFS_I(inode);
2296         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2297             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2298             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2299             nilfs_discontinued(nilfs)) {
2300                 nilfs_transaction_unlock(sb);
2301                 err = nilfs_segctor_sync(sci);
2302                 return err;
2303         }
2304
2305         spin_lock(&nilfs->ns_inode_lock);
2306         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2307             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2308                 spin_unlock(&nilfs->ns_inode_lock);
2309                 nilfs_transaction_unlock(sb);
2310                 return 0;
2311         }
2312         spin_unlock(&nilfs->ns_inode_lock);
2313         sci->sc_dsync_inode = ii;
2314         sci->sc_dsync_start = start;
2315         sci->sc_dsync_end = end;
2316
2317         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2318         if (!err)
2319                 nilfs->ns_flushed_device = 0;
2320
2321         nilfs_transaction_unlock(sb);
2322         return err;
2323 }
2324
2325 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2326 #define FLUSH_DAT_BIT   BIT(NILFS_DAT_INO) /* DAT only */
2327
2328 /**
2329  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2330  * @sci: segment constructor object
2331  */
2332 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2333 {
2334         spin_lock(&sci->sc_state_lock);
2335         sci->sc_seq_accepted = sci->sc_seq_request;
2336         spin_unlock(&sci->sc_state_lock);
2337         del_timer_sync(&sci->sc_timer);
2338 }
2339
2340 /**
2341  * nilfs_segctor_notify - notify the result of request to caller threads
2342  * @sci: segment constructor object
2343  * @mode: mode of log forming
2344  * @err: error code to be notified
2345  */
2346 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2347 {
2348         /* Clear requests (even when the construction failed) */
2349         spin_lock(&sci->sc_state_lock);
2350
2351         if (mode == SC_LSEG_SR) {
2352                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2353                 sci->sc_seq_done = sci->sc_seq_accepted;
2354                 nilfs_segctor_wakeup(sci, err);
2355                 sci->sc_flush_request = 0;
2356         } else {
2357                 if (mode == SC_FLUSH_FILE)
2358                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2359                 else if (mode == SC_FLUSH_DAT)
2360                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2361
2362                 /* re-enable timer if checkpoint creation was not done */
2363                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2364                     time_before(jiffies, sci->sc_timer.expires))
2365                         add_timer(&sci->sc_timer);
2366         }
2367         spin_unlock(&sci->sc_state_lock);
2368 }
2369
2370 /**
2371  * nilfs_segctor_construct - form logs and write them to disk
2372  * @sci: segment constructor object
2373  * @mode: mode of log forming
2374  */
2375 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2376 {
2377         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2378         struct nilfs_super_block **sbp;
2379         int err = 0;
2380
2381         nilfs_segctor_accept(sci);
2382
2383         if (nilfs_discontinued(nilfs))
2384                 mode = SC_LSEG_SR;
2385         if (!nilfs_segctor_confirm(sci))
2386                 err = nilfs_segctor_do_construct(sci, mode);
2387
2388         if (likely(!err)) {
2389                 if (mode != SC_FLUSH_DAT)
2390                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2391                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2392                     nilfs_discontinued(nilfs)) {
2393                         down_write(&nilfs->ns_sem);
2394                         err = -EIO;
2395                         sbp = nilfs_prepare_super(sci->sc_super,
2396                                                   nilfs_sb_will_flip(nilfs));
2397                         if (likely(sbp)) {
2398                                 nilfs_set_log_cursor(sbp[0], nilfs);
2399                                 err = nilfs_commit_super(sci->sc_super,
2400                                                          NILFS_SB_COMMIT);
2401                         }
2402                         up_write(&nilfs->ns_sem);
2403                 }
2404         }
2405
2406         nilfs_segctor_notify(sci, mode, err);
2407         return err;
2408 }
2409
2410 static void nilfs_construction_timeout(struct timer_list *t)
2411 {
2412         struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2413
2414         wake_up_process(sci->sc_timer_task);
2415 }
2416
2417 static void
2418 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2419 {
2420         struct nilfs_inode_info *ii, *n;
2421
2422         list_for_each_entry_safe(ii, n, head, i_dirty) {
2423                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2424                         continue;
2425                 list_del_init(&ii->i_dirty);
2426                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2427                 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2428                 iput(&ii->vfs_inode);
2429         }
2430 }
2431
2432 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2433                          void **kbufs)
2434 {
2435         struct the_nilfs *nilfs = sb->s_fs_info;
2436         struct nilfs_sc_info *sci = nilfs->ns_writer;
2437         struct nilfs_transaction_info ti;
2438         int err;
2439
2440         if (unlikely(!sci))
2441                 return -EROFS;
2442
2443         nilfs_transaction_lock(sb, &ti, 1);
2444
2445         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2446         if (unlikely(err))
2447                 goto out_unlock;
2448
2449         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2450         if (unlikely(err)) {
2451                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2452                 goto out_unlock;
2453         }
2454
2455         sci->sc_freesegs = kbufs[4];
2456         sci->sc_nfreesegs = argv[4].v_nmembs;
2457         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2458
2459         for (;;) {
2460                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2461                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2462
2463                 if (likely(!err))
2464                         break;
2465
2466                 nilfs_warn(sb, "error %d cleaning segments", err);
2467                 set_current_state(TASK_INTERRUPTIBLE);
2468                 schedule_timeout(sci->sc_interval);
2469         }
2470         if (nilfs_test_opt(nilfs, DISCARD)) {
2471                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2472                                                  sci->sc_nfreesegs);
2473                 if (ret) {
2474                         nilfs_warn(sb,
2475                                    "error %d on discard request, turning discards off for the device",
2476                                    ret);
2477                         nilfs_clear_opt(nilfs, DISCARD);
2478                 }
2479         }
2480
2481  out_unlock:
2482         sci->sc_freesegs = NULL;
2483         sci->sc_nfreesegs = 0;
2484         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2485         nilfs_transaction_unlock(sb);
2486         return err;
2487 }
2488
2489 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2490 {
2491         struct nilfs_transaction_info ti;
2492
2493         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2494         nilfs_segctor_construct(sci, mode);
2495
2496         /*
2497          * Unclosed segment should be retried.  We do this using sc_timer.
2498          * Timeout of sc_timer will invoke complete construction which leads
2499          * to close the current logical segment.
2500          */
2501         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2502                 nilfs_segctor_start_timer(sci);
2503
2504         nilfs_transaction_unlock(sci->sc_super);
2505 }
2506
2507 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2508 {
2509         int mode = 0;
2510
2511         spin_lock(&sci->sc_state_lock);
2512         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2513                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2514         spin_unlock(&sci->sc_state_lock);
2515
2516         if (mode) {
2517                 nilfs_segctor_do_construct(sci, mode);
2518
2519                 spin_lock(&sci->sc_state_lock);
2520                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2521                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2522                 spin_unlock(&sci->sc_state_lock);
2523         }
2524         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2525 }
2526
2527 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2528 {
2529         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2530             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2531                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2532                         return SC_FLUSH_FILE;
2533                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2534                         return SC_FLUSH_DAT;
2535         }
2536         return SC_LSEG_SR;
2537 }
2538
2539 /**
2540  * nilfs_segctor_thread - main loop of the segment constructor thread.
2541  * @arg: pointer to a struct nilfs_sc_info.
2542  *
2543  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2544  * to execute segment constructions.
2545  */
2546 static int nilfs_segctor_thread(void *arg)
2547 {
2548         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2549         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2550         int timeout = 0;
2551
2552         sci->sc_timer_task = current;
2553
2554         /* start sync. */
2555         sci->sc_task = current;
2556         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2557         nilfs_info(sci->sc_super,
2558                    "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2559                    sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2560
2561         spin_lock(&sci->sc_state_lock);
2562  loop:
2563         for (;;) {
2564                 int mode;
2565
2566                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2567                         goto end_thread;
2568
2569                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2570                         mode = SC_LSEG_SR;
2571                 else if (sci->sc_flush_request)
2572                         mode = nilfs_segctor_flush_mode(sci);
2573                 else
2574                         break;
2575
2576                 spin_unlock(&sci->sc_state_lock);
2577                 nilfs_segctor_thread_construct(sci, mode);
2578                 spin_lock(&sci->sc_state_lock);
2579                 timeout = 0;
2580         }
2581
2582
2583         if (freezing(current)) {
2584                 spin_unlock(&sci->sc_state_lock);
2585                 try_to_freeze();
2586                 spin_lock(&sci->sc_state_lock);
2587         } else {
2588                 DEFINE_WAIT(wait);
2589                 int should_sleep = 1;
2590
2591                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2592                                 TASK_INTERRUPTIBLE);
2593
2594                 if (sci->sc_seq_request != sci->sc_seq_done)
2595                         should_sleep = 0;
2596                 else if (sci->sc_flush_request)
2597                         should_sleep = 0;
2598                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2599                         should_sleep = time_before(jiffies,
2600                                         sci->sc_timer.expires);
2601
2602                 if (should_sleep) {
2603                         spin_unlock(&sci->sc_state_lock);
2604                         schedule();
2605                         spin_lock(&sci->sc_state_lock);
2606                 }
2607                 finish_wait(&sci->sc_wait_daemon, &wait);
2608                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2609                            time_after_eq(jiffies, sci->sc_timer.expires));
2610
2611                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2612                         set_nilfs_discontinued(nilfs);
2613         }
2614         goto loop;
2615
2616  end_thread:
2617         spin_unlock(&sci->sc_state_lock);
2618
2619         /* end sync. */
2620         sci->sc_task = NULL;
2621         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2622         return 0;
2623 }
2624
2625 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2626 {
2627         struct task_struct *t;
2628
2629         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2630         if (IS_ERR(t)) {
2631                 int err = PTR_ERR(t);
2632
2633                 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2634                           err);
2635                 return err;
2636         }
2637         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2638         return 0;
2639 }
2640
2641 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2642         __acquires(&sci->sc_state_lock)
2643         __releases(&sci->sc_state_lock)
2644 {
2645         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2646
2647         while (sci->sc_task) {
2648                 wake_up(&sci->sc_wait_daemon);
2649                 spin_unlock(&sci->sc_state_lock);
2650                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2651                 spin_lock(&sci->sc_state_lock);
2652         }
2653 }
2654
2655 /*
2656  * Setup & clean-up functions
2657  */
2658 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2659                                                struct nilfs_root *root)
2660 {
2661         struct the_nilfs *nilfs = sb->s_fs_info;
2662         struct nilfs_sc_info *sci;
2663
2664         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2665         if (!sci)
2666                 return NULL;
2667
2668         sci->sc_super = sb;
2669
2670         nilfs_get_root(root);
2671         sci->sc_root = root;
2672
2673         init_waitqueue_head(&sci->sc_wait_request);
2674         init_waitqueue_head(&sci->sc_wait_daemon);
2675         init_waitqueue_head(&sci->sc_wait_task);
2676         spin_lock_init(&sci->sc_state_lock);
2677         INIT_LIST_HEAD(&sci->sc_dirty_files);
2678         INIT_LIST_HEAD(&sci->sc_segbufs);
2679         INIT_LIST_HEAD(&sci->sc_write_logs);
2680         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2681         INIT_LIST_HEAD(&sci->sc_iput_queue);
2682         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2683         timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2684
2685         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2686         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2687         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2688
2689         if (nilfs->ns_interval)
2690                 sci->sc_interval = HZ * nilfs->ns_interval;
2691         if (nilfs->ns_watermark)
2692                 sci->sc_watermark = nilfs->ns_watermark;
2693         return sci;
2694 }
2695
2696 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2697 {
2698         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2699
2700         /*
2701          * The segctord thread was stopped and its timer was removed.
2702          * But some tasks remain.
2703          */
2704         do {
2705                 struct nilfs_transaction_info ti;
2706
2707                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2708                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2709                 nilfs_transaction_unlock(sci->sc_super);
2710
2711                 flush_work(&sci->sc_iput_work);
2712
2713         } while (ret && retrycount-- > 0);
2714 }
2715
2716 /**
2717  * nilfs_segctor_destroy - destroy the segment constructor.
2718  * @sci: nilfs_sc_info
2719  *
2720  * nilfs_segctor_destroy() kills the segctord thread and frees
2721  * the nilfs_sc_info struct.
2722  * Caller must hold the segment semaphore.
2723  */
2724 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2725 {
2726         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2727         int flag;
2728
2729         up_write(&nilfs->ns_segctor_sem);
2730
2731         spin_lock(&sci->sc_state_lock);
2732         nilfs_segctor_kill_thread(sci);
2733         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2734                 || sci->sc_seq_request != sci->sc_seq_done);
2735         spin_unlock(&sci->sc_state_lock);
2736
2737         if (flush_work(&sci->sc_iput_work))
2738                 flag = true;
2739
2740         if (flag || !nilfs_segctor_confirm(sci))
2741                 nilfs_segctor_write_out(sci);
2742
2743         if (!list_empty(&sci->sc_dirty_files)) {
2744                 nilfs_warn(sci->sc_super,
2745                            "disposed unprocessed dirty file(s) when stopping log writer");
2746                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2747         }
2748
2749         if (!list_empty(&sci->sc_iput_queue)) {
2750                 nilfs_warn(sci->sc_super,
2751                            "disposed unprocessed inode(s) in iput queue when stopping log writer");
2752                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2753         }
2754
2755         WARN_ON(!list_empty(&sci->sc_segbufs));
2756         WARN_ON(!list_empty(&sci->sc_write_logs));
2757
2758         nilfs_put_root(sci->sc_root);
2759
2760         down_write(&nilfs->ns_segctor_sem);
2761
2762         del_timer_sync(&sci->sc_timer);
2763         kfree(sci);
2764 }
2765
2766 /**
2767  * nilfs_attach_log_writer - attach log writer
2768  * @sb: super block instance
2769  * @root: root object of the current filesystem tree
2770  *
2771  * This allocates a log writer object, initializes it, and starts the
2772  * log writer.
2773  *
2774  * Return Value: On success, 0 is returned. On error, one of the following
2775  * negative error code is returned.
2776  *
2777  * %-ENOMEM - Insufficient memory available.
2778  */
2779 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2780 {
2781         struct the_nilfs *nilfs = sb->s_fs_info;
2782         int err;
2783
2784         if (nilfs->ns_writer) {
2785                 /*
2786                  * This happens if the filesystem was remounted
2787                  * read/write after nilfs_error degenerated it into a
2788                  * read-only mount.
2789                  */
2790                 nilfs_detach_log_writer(sb);
2791         }
2792
2793         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2794         if (!nilfs->ns_writer)
2795                 return -ENOMEM;
2796
2797         inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2798
2799         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2800         if (unlikely(err))
2801                 nilfs_detach_log_writer(sb);
2802
2803         return err;
2804 }
2805
2806 /**
2807  * nilfs_detach_log_writer - destroy log writer
2808  * @sb: super block instance
2809  *
2810  * This kills log writer daemon, frees the log writer object, and
2811  * destroys list of dirty files.
2812  */
2813 void nilfs_detach_log_writer(struct super_block *sb)
2814 {
2815         struct the_nilfs *nilfs = sb->s_fs_info;
2816         LIST_HEAD(garbage_list);
2817
2818         down_write(&nilfs->ns_segctor_sem);
2819         if (nilfs->ns_writer) {
2820                 nilfs_segctor_destroy(nilfs->ns_writer);
2821                 nilfs->ns_writer = NULL;
2822         }
2823
2824         /* Force to free the list of dirty files */
2825         spin_lock(&nilfs->ns_inode_lock);
2826         if (!list_empty(&nilfs->ns_dirty_files)) {
2827                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2828                 nilfs_warn(sb,
2829                            "disposed unprocessed dirty file(s) when detaching log writer");
2830         }
2831         spin_unlock(&nilfs->ns_inode_lock);
2832         up_write(&nilfs->ns_segctor_sem);
2833
2834         nilfs_dispose_list(nilfs, &garbage_list, 1);
2835 }