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 (sb_rdonly(sb) || unlikely(!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 /**
 439  * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
 440  * @sci: segment constructor object
 441  *
 442  * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
 443  * the current segment summary block.
 444  */
 445 static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
 446 {
 447         struct nilfs_segsum_pointer *ssp;
 448
 449         ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
 450         if (ssp->offset < ssp->bh->b_size)
 451                 memset(ssp->bh->b_data + ssp->offset, 0,
 452                        ssp->bh->b_size - ssp->offset);
 453 }
 454
 455 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 456 {
 457         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 458         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 459                 return -E2BIG; /*
 460                                 * The current segment is filled up
 461                                 * (internal code)
 462                                 */
 463         nilfs_segctor_zeropad_segsum(sci);
 464         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 465         return nilfs_segctor_reset_segment_buffer(sci);
 466 }
 467
 468 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 469 {
 470         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 471         int err;
 472
 473         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 474                 err = nilfs_segctor_feed_segment(sci);
 475                 if (err)
 476                         return err;
 477                 segbuf = sci->sc_curseg;
 478         }
 479         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 480         if (likely(!err))
 481                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 482         return err;
 483 }
 484
 485 /*
 486  * Functions for making segment summary and payloads
 487  */
 488 static int nilfs_segctor_segsum_block_required(
 489         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 490         unsigned int binfo_size)
 491 {
 492         unsigned int blocksize = sci->sc_super->s_blocksize;
 493         /* Size of finfo and binfo is enough small against blocksize */
 494
 495         return ssp->offset + binfo_size +
 496                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 497                 blocksize;
 498 }
 499
 500 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 501                                       struct inode *inode)
 502 {
 503         sci->sc_curseg->sb_sum.nfinfo++;
 504         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 505         nilfs_segctor_map_segsum_entry(
 506                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 507
 508         if (NILFS_I(inode)->i_root &&
 509             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 510                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 511         /* skip finfo */
 512 }
 513
 514 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 515                                     struct inode *inode)
 516 {
 517         struct nilfs_finfo *finfo;
 518         struct nilfs_inode_info *ii;
 519         struct nilfs_segment_buffer *segbuf;
 520         __u64 cno;
 521
 522         if (sci->sc_blk_cnt == 0)
 523                 return;
 524
 525         ii = NILFS_I(inode);
 526
 527         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 528                 cno = ii->i_cno;
 529         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 530                 cno = 0;
 531         else
 532                 cno = sci->sc_cno;
 533
 534         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 535                                                  sizeof(*finfo));
 536         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 537         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 538         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 539         finfo->fi_cno = cpu_to_le64(cno);
 540
 541         segbuf = sci->sc_curseg;
 542         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 543                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 544         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 545         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 546 }
 547
 548 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 549                                         struct buffer_head *bh,
 550                                         struct inode *inode,
 551                                         unsigned int binfo_size)
 552 {
 553         struct nilfs_segment_buffer *segbuf;
 554         int required, err = 0;
 555
 556  retry:
 557         segbuf = sci->sc_curseg;
 558         required = nilfs_segctor_segsum_block_required(
 559                 sci, &sci->sc_binfo_ptr, binfo_size);
 560         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 561                 nilfs_segctor_end_finfo(sci, inode);
 562                 err = nilfs_segctor_feed_segment(sci);
 563                 if (err)
 564                         return err;
 565                 goto retry;
 566         }
 567         if (unlikely(required)) {
 568                 nilfs_segctor_zeropad_segsum(sci);
 569                 err = nilfs_segbuf_extend_segsum(segbuf);
 570                 if (unlikely(err))
 571                         goto failed;
 572         }
 573         if (sci->sc_blk_cnt == 0)
 574                 nilfs_segctor_begin_finfo(sci, inode);
 575
 576         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 577         /* Substitution to vblocknr is delayed until update_blocknr() */
 578         nilfs_segbuf_add_file_buffer(segbuf, bh);
 579         sci->sc_blk_cnt++;
 580  failed:
 581         return err;
 582 }
 583
 584 /*
 585  * Callback functions that enumerate, mark, and collect dirty blocks
 586  */
 587 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 588                                    struct buffer_head *bh, struct inode *inode)
 589 {
 590         int err;
 591
 592         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 593         if (err < 0)
 594                 return err;
 595
 596         err = nilfs_segctor_add_file_block(sci, bh, inode,
 597                                            sizeof(struct nilfs_binfo_v));
 598         if (!err)
 599                 sci->sc_datablk_cnt++;
 600         return err;
 601 }
 602
 603 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 604                                    struct buffer_head *bh,
 605                                    struct inode *inode)
 606 {
 607         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 608 }
 609
 610 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 611                                    struct buffer_head *bh,
 612                                    struct inode *inode)
 613 {
 614         WARN_ON(!buffer_dirty(bh));
 615         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 616 }
 617
 618 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 619                                         struct nilfs_segsum_pointer *ssp,
 620                                         union nilfs_binfo *binfo)
 621 {
 622         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 623                 sci, ssp, sizeof(*binfo_v));
 624         *binfo_v = binfo->bi_v;
 625 }
 626
 627 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 628                                         struct nilfs_segsum_pointer *ssp,
 629                                         union nilfs_binfo *binfo)
 630 {
 631         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 632                 sci, ssp, sizeof(*vblocknr));
 633         *vblocknr = binfo->bi_v.bi_vblocknr;
 634 }
 635
 636 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
 637         .collect_data = nilfs_collect_file_data,
 638         .collect_node = nilfs_collect_file_node,
 639         .collect_bmap = nilfs_collect_file_bmap,
 640         .write_data_binfo = nilfs_write_file_data_binfo,
 641         .write_node_binfo = nilfs_write_file_node_binfo,
 642 };
 643
 644 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 645                                   struct buffer_head *bh, struct inode *inode)
 646 {
 647         int err;
 648
 649         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 650         if (err < 0)
 651                 return err;
 652
 653         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 654         if (!err)
 655                 sci->sc_datablk_cnt++;
 656         return err;
 657 }
 658
 659 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 660                                   struct buffer_head *bh, struct inode *inode)
 661 {
 662         WARN_ON(!buffer_dirty(bh));
 663         return nilfs_segctor_add_file_block(sci, bh, inode,
 664                                             sizeof(struct nilfs_binfo_dat));
 665 }
 666
 667 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 668                                        struct nilfs_segsum_pointer *ssp,
 669                                        union nilfs_binfo *binfo)
 670 {
 671         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 672                                                           sizeof(*blkoff));
 673         *blkoff = binfo->bi_dat.bi_blkoff;
 674 }
 675
 676 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 677                                        struct nilfs_segsum_pointer *ssp,
 678                                        union nilfs_binfo *binfo)
 679 {
 680         struct nilfs_binfo_dat *binfo_dat =
 681                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 682         *binfo_dat = binfo->bi_dat;
 683 }
 684
 685 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
 686         .collect_data = nilfs_collect_dat_data,
 687         .collect_node = nilfs_collect_file_node,
 688         .collect_bmap = nilfs_collect_dat_bmap,
 689         .write_data_binfo = nilfs_write_dat_data_binfo,
 690         .write_node_binfo = nilfs_write_dat_node_binfo,
 691 };
 692
 693 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 694         .collect_data = nilfs_collect_file_data,
 695         .collect_node = NULL,
 696         .collect_bmap = NULL,
 697         .write_data_binfo = nilfs_write_file_data_binfo,
 698         .write_node_binfo = NULL,
 699 };
 700
 701 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 702                                               struct list_head *listp,
 703                                               size_t nlimit,
 704                                               loff_t start, loff_t end)
 705 {
 706         struct address_space *mapping = inode->i_mapping;
 707         struct pagevec pvec;
 708         pgoff_t index = 0, last = ULONG_MAX;
 709         size_t ndirties = 0;
 710         int i;
 711
 712         if (unlikely(start != 0 || end != LLONG_MAX)) {
 713                 /*
 714                  * A valid range is given for sync-ing data pages. The
 715                  * range is rounded to per-page; extra dirty buffers
 716                  * may be included if blocksize < pagesize.
 717                  */
 718                 index = start >> PAGE_SHIFT;
 719                 last = end >> PAGE_SHIFT;
 720         }
 721         pagevec_init(&pvec);
 722  repeat:
 723         if (unlikely(index > last) ||
 724             !pagevec_lookup_range_tag(&pvec, mapping, &index, last,
 725                                 PAGECACHE_TAG_DIRTY))
 726                 return ndirties;
 727
 728         for (i = 0; i < pagevec_count(&pvec); i++) {
 729                 struct buffer_head *bh, *head;
 730                 struct page *page = pvec.pages[i];
 731
 732                 lock_page(page);
 733                 if (unlikely(page->mapping != mapping)) {
 734                         /* Exclude pages removed from the address space */
 735                         unlock_page(page);
 736                         continue;
 737                 }
 738                 if (!page_has_buffers(page))
 739                         create_empty_buffers(page, i_blocksize(inode), 0);
 740                 unlock_page(page);
 741
 742                 bh = head = page_buffers(page);
 743                 do {
 744                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 745                                 continue;
 746                         get_bh(bh);
 747                         list_add_tail(&bh->b_assoc_buffers, listp);
 748                         ndirties++;
 749                         if (unlikely(ndirties >= nlimit)) {
 750                                 pagevec_release(&pvec);
 751                                 cond_resched();
 752                                 return ndirties;
 753                         }
 754                 } while (bh = bh->b_this_page, bh != head);
 755         }
 756         pagevec_release(&pvec);
 757         cond_resched();
 758         goto repeat;
 759 }
 760
 761 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 762                                             struct list_head *listp)
 763 {
 764         struct nilfs_inode_info *ii = NILFS_I(inode);
 765         struct inode *btnc_inode = ii->i_assoc_inode;
 766         struct pagevec pvec;
 767         struct buffer_head *bh, *head;
 768         unsigned int i;
 769         pgoff_t index = 0;
 770
 771         if (!btnc_inode)
 772                 return;
 773
 774         pagevec_init(&pvec);
 775
 776         while (pagevec_lookup_tag(&pvec, btnc_inode->i_mapping, &index,
 777                                         PAGECACHE_TAG_DIRTY)) {
 778                 for (i = 0; i < pagevec_count(&pvec); i++) {
 779                         bh = head = page_buffers(pvec.pages[i]);
 780                         do {
 781                                 if (buffer_dirty(bh) &&
 782                                                 !buffer_async_write(bh)) {
 783                                         get_bh(bh);
 784                                         list_add_tail(&bh->b_assoc_buffers,
 785                                                       listp);
 786                                 }
 787                                 bh = bh->b_this_page;
 788                         } while (bh != head);
 789                 }
 790                 pagevec_release(&pvec);
 791                 cond_resched();
 792         }
 793 }
 794
 795 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 796                                struct list_head *head, int force)
 797 {
 798         struct nilfs_inode_info *ii, *n;
 799         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 800         unsigned int nv = 0;
 801
 802         while (!list_empty(head)) {
 803                 spin_lock(&nilfs->ns_inode_lock);
 804                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 805                         list_del_init(&ii->i_dirty);
 806                         if (force) {
 807                                 if (unlikely(ii->i_bh)) {
 808                                         brelse(ii->i_bh);
 809                                         ii->i_bh = NULL;
 810                                 }
 811                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 812                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 813                                 list_add_tail(&ii->i_dirty,
 814                                               &nilfs->ns_dirty_files);
 815                                 continue;
 816                         }
 817                         ivec[nv++] = ii;
 818                         if (nv == SC_N_INODEVEC)
 819                                 break;
 820                 }
 821                 spin_unlock(&nilfs->ns_inode_lock);
 822
 823                 for (pii = ivec; nv > 0; pii++, nv--)
 824                         iput(&(*pii)->vfs_inode);
 825         }
 826 }
 827
 828 static void nilfs_iput_work_func(struct work_struct *work)
 829 {
 830         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 831                                                  sc_iput_work);
 832         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 833
 834         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 835 }
 836
 837 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 838                                      struct nilfs_root *root)
 839 {
 840         int ret = 0;
 841
 842         if (nilfs_mdt_fetch_dirty(root->ifile))
 843                 ret++;
 844         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 845                 ret++;
 846         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 847                 ret++;
 848         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 849                 ret++;
 850         return ret;
 851 }
 852
 853 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 854 {
 855         return list_empty(&sci->sc_dirty_files) &&
 856                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 857                 sci->sc_nfreesegs == 0 &&
 858                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 859 }
 860
 861 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 862 {
 863         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 864         int ret = 0;
 865
 866         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 867                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 868
 869         spin_lock(&nilfs->ns_inode_lock);
 870         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 871                 ret++;
 872
 873         spin_unlock(&nilfs->ns_inode_lock);
 874         return ret;
 875 }
 876
 877 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 878 {
 879         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 880
 881         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 882         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 883         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 884         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 885 }
 886
 887 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 888 {
 889         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 890         struct buffer_head *bh_cp;
 891         struct nilfs_checkpoint *raw_cp;
 892         int err;
 893
 894         /* XXX: this interface will be changed */
 895         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 896                                           &raw_cp, &bh_cp);
 897         if (likely(!err)) {
 898                 /*
 899                  * The following code is duplicated with cpfile.  But, it is
 900                  * needed to collect the checkpoint even if it was not newly
 901                  * created.
 902                  */
 903                 mark_buffer_dirty(bh_cp);
 904                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 905                 nilfs_cpfile_put_checkpoint(
 906                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 907         } else if (err == -EINVAL || err == -ENOENT) {
 908                 nilfs_error(sci->sc_super,
 909                             "checkpoint creation failed due to metadata corruption.");
 910                 err = -EIO;
 911         }
 912         return err;
 913 }
 914
 915 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 916 {
 917         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 918         struct buffer_head *bh_cp;
 919         struct nilfs_checkpoint *raw_cp;
 920         int err;
 921
 922         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 923                                           &raw_cp, &bh_cp);
 924         if (unlikely(err)) {
 925                 if (err == -EINVAL || err == -ENOENT) {
 926                         nilfs_error(sci->sc_super,
 927                                     "checkpoint finalization failed due to metadata corruption.");
 928                         err = -EIO;
 929                 }
 930                 goto failed_ibh;
 931         }
 932         raw_cp->cp_snapshot_list.ssl_next = 0;
 933         raw_cp->cp_snapshot_list.ssl_prev = 0;
 934         raw_cp->cp_inodes_count =
 935                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 936         raw_cp->cp_blocks_count =
 937                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 938         raw_cp->cp_nblk_inc =
 939                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 940         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 941         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 942
 943         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 944                 nilfs_checkpoint_clear_minor(raw_cp);
 945         else
 946                 nilfs_checkpoint_set_minor(raw_cp);
 947
 948         nilfs_write_inode_common(sci->sc_root->ifile,
 949                                  &raw_cp->cp_ifile_inode, 1);
 950         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 951         return 0;
 952
 953  failed_ibh:
 954         return err;
 955 }
 956
 957 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 958                                     struct nilfs_inode_info *ii)
 959
 960 {
 961         struct buffer_head *ibh;
 962         struct nilfs_inode *raw_inode;
 963
 964         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 965                 ibh = ii->i_bh;
 966                 BUG_ON(!ibh);
 967                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 968                                                   ibh);
 969                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 970                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 971         }
 972 }
 973
 974 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 975 {
 976         struct nilfs_inode_info *ii;
 977
 978         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 979                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 980                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 981         }
 982 }
 983
 984 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 985                                              struct the_nilfs *nilfs)
 986 {
 987         struct buffer_head *bh_sr;
 988         struct nilfs_super_root *raw_sr;
 989         unsigned int isz, srsz;
 990
 991         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 992
 993         lock_buffer(bh_sr);
 994         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 995         isz = nilfs->ns_inode_size;
 996         srsz = NILFS_SR_BYTES(isz);
 997
 998         raw_sr->sr_sum = 0;  /* Ensure initialization within this update */
 999         raw_sr->sr_bytes = cpu_to_le16(srsz);
1000         raw_sr->sr_nongc_ctime
1001                 = cpu_to_le64(nilfs_doing_gc() ?
1002                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
1003         raw_sr->sr_flags = 0;
1004
1005         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
1006                                  NILFS_SR_DAT_OFFSET(isz), 1);
1007         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
1008                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
1009         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
1010                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
1011         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1012         set_buffer_uptodate(bh_sr);
1013         unlock_buffer(bh_sr);
1014 }
1015
1016 static void nilfs_redirty_inodes(struct list_head *head)
1017 {
1018         struct nilfs_inode_info *ii;
1019
1020         list_for_each_entry(ii, head, i_dirty) {
1021                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1022                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1023         }
1024 }
1025
1026 static void nilfs_drop_collected_inodes(struct list_head *head)
1027 {
1028         struct nilfs_inode_info *ii;
1029
1030         list_for_each_entry(ii, head, i_dirty) {
1031                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1032                         continue;
1033
1034                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1035                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1036         }
1037 }
1038
1039 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1040                                        struct inode *inode,
1041                                        struct list_head *listp,
1042                                        int (*collect)(struct nilfs_sc_info *,
1043                                                       struct buffer_head *,
1044                                                       struct inode *))
1045 {
1046         struct buffer_head *bh, *n;
1047         int err = 0;
1048
1049         if (collect) {
1050                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1051                         list_del_init(&bh->b_assoc_buffers);
1052                         err = collect(sci, bh, inode);
1053                         brelse(bh);
1054                         if (unlikely(err))
1055                                 goto dispose_buffers;
1056                 }
1057                 return 0;
1058         }
1059
1060  dispose_buffers:
1061         while (!list_empty(listp)) {
1062                 bh = list_first_entry(listp, struct buffer_head,
1063                                       b_assoc_buffers);
1064                 list_del_init(&bh->b_assoc_buffers);
1065                 brelse(bh);
1066         }
1067         return err;
1068 }
1069
1070 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1071 {
1072         /* Remaining number of blocks within segment buffer */
1073         return sci->sc_segbuf_nblocks -
1074                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1075 }
1076
1077 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1078                                    struct inode *inode,
1079                                    const struct nilfs_sc_operations *sc_ops)
1080 {
1081         LIST_HEAD(data_buffers);
1082         LIST_HEAD(node_buffers);
1083         int err;
1084
1085         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1086                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1087
1088                 n = nilfs_lookup_dirty_data_buffers(
1089                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1090                 if (n > rest) {
1091                         err = nilfs_segctor_apply_buffers(
1092                                 sci, inode, &data_buffers,
1093                                 sc_ops->collect_data);
1094                         BUG_ON(!err); /* always receive -E2BIG or true error */
1095                         goto break_or_fail;
1096                 }
1097         }
1098         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1099
1100         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1101                 err = nilfs_segctor_apply_buffers(
1102                         sci, inode, &data_buffers, sc_ops->collect_data);
1103                 if (unlikely(err)) {
1104                         /* dispose node list */
1105                         nilfs_segctor_apply_buffers(
1106                                 sci, inode, &node_buffers, NULL);
1107                         goto break_or_fail;
1108                 }
1109                 sci->sc_stage.flags |= NILFS_CF_NODE;
1110         }
1111         /* Collect node */
1112         err = nilfs_segctor_apply_buffers(
1113                 sci, inode, &node_buffers, sc_ops->collect_node);
1114         if (unlikely(err))
1115                 goto break_or_fail;
1116
1117         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1118         err = nilfs_segctor_apply_buffers(
1119                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1120         if (unlikely(err))
1121                 goto break_or_fail;
1122
1123         nilfs_segctor_end_finfo(sci, inode);
1124         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1125
1126  break_or_fail:
1127         return err;
1128 }
1129
1130 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1131                                          struct inode *inode)
1132 {
1133         LIST_HEAD(data_buffers);
1134         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1135         int err;
1136
1137         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1138                                             sci->sc_dsync_start,
1139                                             sci->sc_dsync_end);
1140
1141         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1142                                           nilfs_collect_file_data);
1143         if (!err) {
1144                 nilfs_segctor_end_finfo(sci, inode);
1145                 BUG_ON(n > rest);
1146                 /* always receive -E2BIG or true error if n > rest */
1147         }
1148         return err;
1149 }
1150
1151 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1152 {
1153         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1154         struct list_head *head;
1155         struct nilfs_inode_info *ii;
1156         size_t ndone;
1157         int err = 0;
1158
1159         switch (nilfs_sc_cstage_get(sci)) {
1160         case NILFS_ST_INIT:
1161                 /* Pre-processes */
1162                 sci->sc_stage.flags = 0;
1163
1164                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1165                         sci->sc_nblk_inc = 0;
1166                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1167                         if (mode == SC_LSEG_DSYNC) {
1168                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1169                                 goto dsync_mode;
1170                         }
1171                 }
1172
1173                 sci->sc_stage.dirty_file_ptr = NULL;
1174                 sci->sc_stage.gc_inode_ptr = NULL;
1175                 if (mode == SC_FLUSH_DAT) {
1176                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1177                         goto dat_stage;
1178                 }
1179                 nilfs_sc_cstage_inc(sci);
1180                 fallthrough;
1181         case NILFS_ST_GC:
1182                 if (nilfs_doing_gc()) {
1183                         head = &sci->sc_gc_inodes;
1184                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1185                                                 head, i_dirty);
1186                         list_for_each_entry_continue(ii, head, i_dirty) {
1187                                 err = nilfs_segctor_scan_file(
1188                                         sci, &ii->vfs_inode,
1189                                         &nilfs_sc_file_ops);
1190                                 if (unlikely(err)) {
1191                                         sci->sc_stage.gc_inode_ptr = list_entry(
1192                                                 ii->i_dirty.prev,
1193                                                 struct nilfs_inode_info,
1194                                                 i_dirty);
1195                                         goto break_or_fail;
1196                                 }
1197                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1198                         }
1199                         sci->sc_stage.gc_inode_ptr = NULL;
1200                 }
1201                 nilfs_sc_cstage_inc(sci);
1202                 fallthrough;
1203         case NILFS_ST_FILE:
1204                 head = &sci->sc_dirty_files;
1205                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1206                                         i_dirty);
1207                 list_for_each_entry_continue(ii, head, i_dirty) {
1208                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1209
1210                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1211                                                       &nilfs_sc_file_ops);
1212                         if (unlikely(err)) {
1213                                 sci->sc_stage.dirty_file_ptr =
1214                                         list_entry(ii->i_dirty.prev,
1215                                                    struct nilfs_inode_info,
1216                                                    i_dirty);
1217                                 goto break_or_fail;
1218                         }
1219                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1220                         /* XXX: required ? */
1221                 }
1222                 sci->sc_stage.dirty_file_ptr = NULL;
1223                 if (mode == SC_FLUSH_FILE) {
1224                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1225                         return 0;
1226                 }
1227                 nilfs_sc_cstage_inc(sci);
1228                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1229                 fallthrough;
1230         case NILFS_ST_IFILE:
1231                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1232                                               &nilfs_sc_file_ops);
1233                 if (unlikely(err))
1234                         break;
1235                 nilfs_sc_cstage_inc(sci);
1236                 /* Creating a checkpoint */
1237                 err = nilfs_segctor_create_checkpoint(sci);
1238                 if (unlikely(err))
1239                         break;
1240                 fallthrough;
1241         case NILFS_ST_CPFILE:
1242                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1243                                               &nilfs_sc_file_ops);
1244                 if (unlikely(err))
1245                         break;
1246                 nilfs_sc_cstage_inc(sci);
1247                 fallthrough;
1248         case NILFS_ST_SUFILE:
1249                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1250                                          sci->sc_nfreesegs, &ndone);
1251                 if (unlikely(err)) {
1252                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1253                                                   sci->sc_freesegs, ndone,
1254                                                   NULL);
1255                         break;
1256                 }
1257                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1258
1259                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1260                                               &nilfs_sc_file_ops);
1261                 if (unlikely(err))
1262                         break;
1263                 nilfs_sc_cstage_inc(sci);
1264                 fallthrough;
1265         case NILFS_ST_DAT:
1266  dat_stage:
1267                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1268                                               &nilfs_sc_dat_ops);
1269                 if (unlikely(err))
1270                         break;
1271                 if (mode == SC_FLUSH_DAT) {
1272                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1273                         return 0;
1274                 }
1275                 nilfs_sc_cstage_inc(sci);
1276                 fallthrough;
1277         case NILFS_ST_SR:
1278                 if (mode == SC_LSEG_SR) {
1279                         /* Appending a super root */
1280                         err = nilfs_segctor_add_super_root(sci);
1281                         if (unlikely(err))
1282                                 break;
1283                 }
1284                 /* End of a logical segment */
1285                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1286                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1287                 return 0;
1288         case NILFS_ST_DSYNC:
1289  dsync_mode:
1290                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1291                 ii = sci->sc_dsync_inode;
1292                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1293                         break;
1294
1295                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1296                 if (unlikely(err))
1297                         break;
1298                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1299                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1300                 return 0;
1301         case NILFS_ST_DONE:
1302                 return 0;
1303         default:
1304                 BUG();
1305         }
1306
1307  break_or_fail:
1308         return err;
1309 }
1310
1311 /**
1312  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1313  * @sci: nilfs_sc_info
1314  * @nilfs: nilfs object
1315  */
1316 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1317                                             struct the_nilfs *nilfs)
1318 {
1319         struct nilfs_segment_buffer *segbuf, *prev;
1320         __u64 nextnum;
1321         int err, alloc = 0;
1322
1323         segbuf = nilfs_segbuf_new(sci->sc_super);
1324         if (unlikely(!segbuf))
1325                 return -ENOMEM;
1326
1327         if (list_empty(&sci->sc_write_logs)) {
1328                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1329                                  nilfs->ns_pseg_offset, nilfs);
1330                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1331                         nilfs_shift_to_next_segment(nilfs);
1332                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1333                 }
1334
1335                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1336                 nextnum = nilfs->ns_nextnum;
1337
1338                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1339                         /* Start from the head of a new full segment */
1340                         alloc++;
1341         } else {
1342                 /* Continue logs */
1343                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1344                 nilfs_segbuf_map_cont(segbuf, prev);
1345                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1346                 nextnum = prev->sb_nextnum;
1347
1348                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1349                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1350                         segbuf->sb_sum.seg_seq++;
1351                         alloc++;
1352                 }
1353         }
1354
1355         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1356         if (err)
1357                 goto failed;
1358
1359         if (alloc) {
1360                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1361                 if (err)
1362                         goto failed;
1363         }
1364         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1365
1366         BUG_ON(!list_empty(&sci->sc_segbufs));
1367         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1368         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1369         return 0;
1370
1371  failed:
1372         nilfs_segbuf_free(segbuf);
1373         return err;
1374 }
1375
1376 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1377                                          struct the_nilfs *nilfs, int nadd)
1378 {
1379         struct nilfs_segment_buffer *segbuf, *prev;
1380         struct inode *sufile = nilfs->ns_sufile;
1381         __u64 nextnextnum;
1382         LIST_HEAD(list);
1383         int err, ret, i;
1384
1385         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1386         /*
1387          * Since the segment specified with nextnum might be allocated during
1388          * the previous construction, the buffer including its segusage may
1389          * not be dirty.  The following call ensures that the buffer is dirty
1390          * and will pin the buffer on memory until the sufile is written.
1391          */
1392         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1393         if (unlikely(err))
1394                 return err;
1395
1396         for (i = 0; i < nadd; i++) {
1397                 /* extend segment info */
1398                 err = -ENOMEM;
1399                 segbuf = nilfs_segbuf_new(sci->sc_super);
1400                 if (unlikely(!segbuf))
1401                         goto failed;
1402
1403                 /* map this buffer to region of segment on-disk */
1404                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1405                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1406
1407                 /* allocate the next next full segment */
1408                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1409                 if (unlikely(err))
1410                         goto failed_segbuf;
1411
1412                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1413                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1414
1415                 list_add_tail(&segbuf->sb_list, &list);
1416                 prev = segbuf;
1417         }
1418         list_splice_tail(&list, &sci->sc_segbufs);
1419         return 0;
1420
1421  failed_segbuf:
1422         nilfs_segbuf_free(segbuf);
1423  failed:
1424         list_for_each_entry(segbuf, &list, sb_list) {
1425                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1426                 WARN_ON(ret); /* never fails */
1427         }
1428         nilfs_destroy_logs(&list);
1429         return err;
1430 }
1431
1432 static void nilfs_free_incomplete_logs(struct list_head *logs,
1433                                        struct the_nilfs *nilfs)
1434 {
1435         struct nilfs_segment_buffer *segbuf, *prev;
1436         struct inode *sufile = nilfs->ns_sufile;
1437         int ret;
1438
1439         segbuf = NILFS_FIRST_SEGBUF(logs);
1440         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1441                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1442                 WARN_ON(ret); /* never fails */
1443         }
1444         if (atomic_read(&segbuf->sb_err)) {
1445                 /* Case 1: The first segment failed */
1446                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1447                         /*
1448                          * Case 1a:  Partial segment appended into an existing
1449                          * segment
1450                          */
1451                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1452                                                 segbuf->sb_fseg_end);
1453                 else /* Case 1b:  New full segment */
1454                         set_nilfs_discontinued(nilfs);
1455         }
1456
1457         prev = segbuf;
1458         list_for_each_entry_continue(segbuf, logs, sb_list) {
1459                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1460                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1461                         WARN_ON(ret); /* never fails */
1462                 }
1463                 if (atomic_read(&segbuf->sb_err) &&
1464                     segbuf->sb_segnum != nilfs->ns_nextnum)
1465                         /* Case 2: extended segment (!= next) failed */
1466                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1467                 prev = segbuf;
1468         }
1469 }
1470
1471 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1472                                           struct inode *sufile)
1473 {
1474         struct nilfs_segment_buffer *segbuf;
1475         unsigned long live_blocks;
1476         int ret;
1477
1478         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1479                 live_blocks = segbuf->sb_sum.nblocks +
1480                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1481                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1482                                                      live_blocks,
1483                                                      sci->sc_seg_ctime);
1484                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1485         }
1486 }
1487
1488 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1489 {
1490         struct nilfs_segment_buffer *segbuf;
1491         int ret;
1492
1493         segbuf = NILFS_FIRST_SEGBUF(logs);
1494         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1495                                              segbuf->sb_pseg_start -
1496                                              segbuf->sb_fseg_start, 0);
1497         WARN_ON(ret); /* always succeed because the segusage is dirty */
1498
1499         list_for_each_entry_continue(segbuf, logs, sb_list) {
1500                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1501                                                      0, 0);
1502                 WARN_ON(ret); /* always succeed */
1503         }
1504 }
1505
1506 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1507                                             struct nilfs_segment_buffer *last,
1508                                             struct inode *sufile)
1509 {
1510         struct nilfs_segment_buffer *segbuf = last;
1511         int ret;
1512
1513         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1514                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1515                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1516                 WARN_ON(ret);
1517         }
1518         nilfs_truncate_logs(&sci->sc_segbufs, last);
1519 }
1520
1521
1522 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1523                                  struct the_nilfs *nilfs, int mode)
1524 {
1525         struct nilfs_cstage prev_stage = sci->sc_stage;
1526         int err, nadd = 1;
1527
1528         /* Collection retry loop */
1529         for (;;) {
1530                 sci->sc_nblk_this_inc = 0;
1531                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1532
1533                 err = nilfs_segctor_reset_segment_buffer(sci);
1534                 if (unlikely(err))
1535                         goto failed;
1536
1537                 err = nilfs_segctor_collect_blocks(sci, mode);
1538                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1539                 if (!err)
1540                         break;
1541
1542                 if (unlikely(err != -E2BIG))
1543                         goto failed;
1544
1545                 /* The current segment is filled up */
1546                 if (mode != SC_LSEG_SR ||
1547                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1548                         break;
1549
1550                 nilfs_clear_logs(&sci->sc_segbufs);
1551
1552                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1553                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1554                                                         sci->sc_freesegs,
1555                                                         sci->sc_nfreesegs,
1556                                                         NULL);
1557                         WARN_ON(err); /* do not happen */
1558                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1559                 }
1560
1561                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1562                 if (unlikely(err))
1563                         return err;
1564
1565                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1566                 sci->sc_stage = prev_stage;
1567         }
1568         nilfs_segctor_zeropad_segsum(sci);
1569         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1570         return 0;
1571
1572  failed:
1573         return err;
1574 }
1575
1576 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1577                                       struct buffer_head *new_bh)
1578 {
1579         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1580
1581         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1582         /* The caller must release old_bh */
1583 }
1584
1585 static int
1586 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1587                                      struct nilfs_segment_buffer *segbuf,
1588                                      int mode)
1589 {
1590         struct inode *inode = NULL;
1591         sector_t blocknr;
1592         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1593         unsigned long nblocks = 0, ndatablk = 0;
1594         const struct nilfs_sc_operations *sc_op = NULL;
1595         struct nilfs_segsum_pointer ssp;
1596         struct nilfs_finfo *finfo = NULL;
1597         union nilfs_binfo binfo;
1598         struct buffer_head *bh, *bh_org;
1599         ino_t ino = 0;
1600         int err = 0;
1601
1602         if (!nfinfo)
1603                 goto out;
1604
1605         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1606         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1607         ssp.offset = sizeof(struct nilfs_segment_summary);
1608
1609         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1610                 if (bh == segbuf->sb_super_root)
1611                         break;
1612                 if (!finfo) {
1613                         finfo = nilfs_segctor_map_segsum_entry(
1614                                 sci, &ssp, sizeof(*finfo));
1615                         ino = le64_to_cpu(finfo->fi_ino);
1616                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1617                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1618
1619                         inode = bh->b_page->mapping->host;
1620
1621                         if (mode == SC_LSEG_DSYNC)
1622                                 sc_op = &nilfs_sc_dsync_ops;
1623                         else if (ino == NILFS_DAT_INO)
1624                                 sc_op = &nilfs_sc_dat_ops;
1625                         else /* file blocks */
1626                                 sc_op = &nilfs_sc_file_ops;
1627                 }
1628                 bh_org = bh;
1629                 get_bh(bh_org);
1630                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1631                                         &binfo);
1632                 if (bh != bh_org)
1633                         nilfs_list_replace_buffer(bh_org, bh);
1634                 brelse(bh_org);
1635                 if (unlikely(err))
1636                         goto failed_bmap;
1637
1638                 if (ndatablk > 0)
1639                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1640                 else
1641                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1642
1643                 blocknr++;
1644                 if (--nblocks == 0) {
1645                         finfo = NULL;
1646                         if (--nfinfo == 0)
1647                                 break;
1648                 } else if (ndatablk > 0)
1649                         ndatablk--;
1650         }
1651  out:
1652         return 0;
1653
1654  failed_bmap:
1655         return err;
1656 }
1657
1658 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1659 {
1660         struct nilfs_segment_buffer *segbuf;
1661         int err;
1662
1663         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1664                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1665                 if (unlikely(err))
1666                         return err;
1667                 nilfs_segbuf_fill_in_segsum(segbuf);
1668         }
1669         return 0;
1670 }
1671
1672 static void nilfs_begin_page_io(struct page *page)
1673 {
1674         if (!page || PageWriteback(page))
1675                 /*
1676                  * For split b-tree node pages, this function may be called
1677                  * twice.  We ignore the 2nd or later calls by this check.
1678                  */
1679                 return;
1680
1681         lock_page(page);
1682         clear_page_dirty_for_io(page);
1683         set_page_writeback(page);
1684         unlock_page(page);
1685 }
1686
1687 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1688 {
1689         struct nilfs_segment_buffer *segbuf;
1690         struct page *bd_page = NULL, *fs_page = NULL;
1691
1692         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1693                 struct buffer_head *bh;
1694
1695                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1696                                     b_assoc_buffers) {
1697                         if (bh->b_page != bd_page) {
1698                                 if (bd_page) {
1699                                         lock_page(bd_page);
1700                                         clear_page_dirty_for_io(bd_page);
1701                                         set_page_writeback(bd_page);
1702                                         unlock_page(bd_page);
1703                                 }
1704                                 bd_page = bh->b_page;
1705                         }
1706                 }
1707
1708                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1709                                     b_assoc_buffers) {
1710                         if (bh == segbuf->sb_super_root) {
1711                                 if (bh->b_page != bd_page) {
1712                                         lock_page(bd_page);
1713                                         clear_page_dirty_for_io(bd_page);
1714                                         set_page_writeback(bd_page);
1715                                         unlock_page(bd_page);
1716                                         bd_page = bh->b_page;
1717                                 }
1718                                 break;
1719                         }
1720                         set_buffer_async_write(bh);
1721                         if (bh->b_page != fs_page) {
1722                                 nilfs_begin_page_io(fs_page);
1723                                 fs_page = bh->b_page;
1724                         }
1725                 }
1726         }
1727         if (bd_page) {
1728                 lock_page(bd_page);
1729                 clear_page_dirty_for_io(bd_page);
1730                 set_page_writeback(bd_page);
1731                 unlock_page(bd_page);
1732         }
1733         nilfs_begin_page_io(fs_page);
1734 }
1735
1736 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1737                                struct the_nilfs *nilfs)
1738 {
1739         int ret;
1740
1741         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1742         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1743         return ret;
1744 }
1745
1746 static void nilfs_end_page_io(struct page *page, int err)
1747 {
1748         if (!page)
1749                 return;
1750
1751         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1752                 /*
1753                  * For b-tree node pages, this function may be called twice
1754                  * or more because they might be split in a segment.
1755                  */
1756                 if (PageDirty(page)) {
1757                         /*
1758                          * For pages holding split b-tree node buffers, dirty
1759                          * flag on the buffers may be cleared discretely.
1760                          * In that case, the page is once redirtied for
1761                          * remaining buffers, and it must be cancelled if
1762                          * all the buffers get cleaned later.
1763                          */
1764                         lock_page(page);
1765                         if (nilfs_page_buffers_clean(page))
1766                                 __nilfs_clear_page_dirty(page);
1767                         unlock_page(page);
1768                 }
1769                 return;
1770         }
1771
1772         if (!err) {
1773                 if (!nilfs_page_buffers_clean(page))
1774                         __set_page_dirty_nobuffers(page);
1775                 ClearPageError(page);
1776         } else {
1777                 __set_page_dirty_nobuffers(page);
1778                 SetPageError(page);
1779         }
1780
1781         end_page_writeback(page);
1782 }
1783
1784 static void nilfs_abort_logs(struct list_head *logs, int err)
1785 {
1786         struct nilfs_segment_buffer *segbuf;
1787         struct page *bd_page = NULL, *fs_page = NULL;
1788         struct buffer_head *bh;
1789
1790         if (list_empty(logs))
1791                 return;
1792
1793         list_for_each_entry(segbuf, logs, sb_list) {
1794                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1795                                     b_assoc_buffers) {
1796                         clear_buffer_uptodate(bh);
1797                         if (bh->b_page != bd_page) {
1798                                 if (bd_page)
1799                                         end_page_writeback(bd_page);
1800                                 bd_page = bh->b_page;
1801                         }
1802                 }
1803
1804                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1805                                     b_assoc_buffers) {
1806                         if (bh == segbuf->sb_super_root) {
1807                                 clear_buffer_uptodate(bh);
1808                                 if (bh->b_page != bd_page) {
1809                                         end_page_writeback(bd_page);
1810                                         bd_page = bh->b_page;
1811                                 }
1812                                 break;
1813                         }
1814                         clear_buffer_async_write(bh);
1815                         if (bh->b_page != fs_page) {
1816                                 nilfs_end_page_io(fs_page, err);
1817                                 fs_page = bh->b_page;
1818                         }
1819                 }
1820         }
1821         if (bd_page)
1822                 end_page_writeback(bd_page);
1823
1824         nilfs_end_page_io(fs_page, err);
1825 }
1826
1827 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1828                                              struct the_nilfs *nilfs, int err)
1829 {
1830         LIST_HEAD(logs);
1831         int ret;
1832
1833         list_splice_tail_init(&sci->sc_write_logs, &logs);
1834         ret = nilfs_wait_on_logs(&logs);
1835         nilfs_abort_logs(&logs, ret ? : err);
1836
1837         list_splice_tail_init(&sci->sc_segbufs, &logs);
1838         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1839         nilfs_free_incomplete_logs(&logs, nilfs);
1840
1841         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1842                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1843                                                 sci->sc_freesegs,
1844                                                 sci->sc_nfreesegs,
1845                                                 NULL);
1846                 WARN_ON(ret); /* do not happen */
1847         }
1848
1849         nilfs_destroy_logs(&logs);
1850 }
1851
1852 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1853                                    struct nilfs_segment_buffer *segbuf)
1854 {
1855         nilfs->ns_segnum = segbuf->sb_segnum;
1856         nilfs->ns_nextnum = segbuf->sb_nextnum;
1857         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1858                 + segbuf->sb_sum.nblocks;
1859         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1860         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1861 }
1862
1863 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1864 {
1865         struct nilfs_segment_buffer *segbuf;
1866         struct page *bd_page = NULL, *fs_page = NULL;
1867         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1868         int update_sr = false;
1869
1870         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1871                 struct buffer_head *bh;
1872
1873                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1874                                     b_assoc_buffers) {
1875                         set_buffer_uptodate(bh);
1876                         clear_buffer_dirty(bh);
1877                         if (bh->b_page != bd_page) {
1878                                 if (bd_page)
1879                                         end_page_writeback(bd_page);
1880                                 bd_page = bh->b_page;
1881                         }
1882                 }
1883                 /*
1884                  * We assume that the buffers which belong to the same page
1885                  * continue over the buffer list.
1886                  * Under this assumption, the last BHs of pages is
1887                  * identifiable by the discontinuity of bh->b_page
1888                  * (page != fs_page).
1889                  *
1890                  * For B-tree node blocks, however, this assumption is not
1891                  * guaranteed.  The cleanup code of B-tree node pages needs
1892                  * special care.
1893                  */
1894                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1895                                     b_assoc_buffers) {
1896                         const unsigned long set_bits = BIT(BH_Uptodate);
1897                         const unsigned long clear_bits =
1898                                 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1899                                  BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1900                                  BIT(BH_NILFS_Redirected));
1901
1902                         if (bh == segbuf->sb_super_root) {
1903                                 set_buffer_uptodate(bh);
1904                                 clear_buffer_dirty(bh);
1905                                 if (bh->b_page != bd_page) {
1906                                         end_page_writeback(bd_page);
1907                                         bd_page = bh->b_page;
1908                                 }
1909                                 update_sr = true;
1910                                 break;
1911                         }
1912                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1913                         if (bh->b_page != fs_page) {
1914                                 nilfs_end_page_io(fs_page, 0);
1915                                 fs_page = bh->b_page;
1916                         }
1917                 }
1918
1919                 if (!nilfs_segbuf_simplex(segbuf)) {
1920                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1921                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1922                                 sci->sc_lseg_stime = jiffies;
1923                         }
1924                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1925                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1926                 }
1927         }
1928         /*
1929          * Since pages may continue over multiple segment buffers,
1930          * end of the last page must be checked outside of the loop.
1931          */
1932         if (bd_page)
1933                 end_page_writeback(bd_page);
1934
1935         nilfs_end_page_io(fs_page, 0);
1936
1937         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1938
1939         if (nilfs_doing_gc())
1940                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1941         else
1942                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1943
1944         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1945
1946         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1947         nilfs_set_next_segment(nilfs, segbuf);
1948
1949         if (update_sr) {
1950                 nilfs->ns_flushed_device = 0;
1951                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1952                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1953
1954                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1955                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1956                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1957                 nilfs_segctor_clear_metadata_dirty(sci);
1958         } else
1959                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1960 }
1961
1962 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1963 {
1964         int ret;
1965
1966         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1967         if (!ret) {
1968                 nilfs_segctor_complete_write(sci);
1969                 nilfs_destroy_logs(&sci->sc_write_logs);
1970         }
1971         return ret;
1972 }
1973
1974 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1975                                              struct the_nilfs *nilfs)
1976 {
1977         struct nilfs_inode_info *ii, *n;
1978         struct inode *ifile = sci->sc_root->ifile;
1979
1980         spin_lock(&nilfs->ns_inode_lock);
1981  retry:
1982         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1983                 if (!ii->i_bh) {
1984                         struct buffer_head *ibh;
1985                         int err;
1986
1987                         spin_unlock(&nilfs->ns_inode_lock);
1988                         err = nilfs_ifile_get_inode_block(
1989                                 ifile, ii->vfs_inode.i_ino, &ibh);
1990                         if (unlikely(err)) {
1991                                 nilfs_warn(sci->sc_super,
1992                                            "log writer: error %d getting inode block (ino=%lu)",
1993                                            err, ii->vfs_inode.i_ino);
1994                                 return err;
1995                         }
1996                         spin_lock(&nilfs->ns_inode_lock);
1997                         if (likely(!ii->i_bh))
1998                                 ii->i_bh = ibh;
1999                         else
2000                                 brelse(ibh);
2001                         goto retry;
2002                 }
2003
2004                 // Always redirty the buffer to avoid race condition
2005                 mark_buffer_dirty(ii->i_bh);
2006                 nilfs_mdt_mark_dirty(ifile);
2007
2008                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2009                 set_bit(NILFS_I_BUSY, &ii->i_state);
2010                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2011         }
2012         spin_unlock(&nilfs->ns_inode_lock);
2013
2014         return 0;
2015 }
2016
2017 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2018                                              struct the_nilfs *nilfs)
2019 {
2020         struct nilfs_inode_info *ii, *n;
2021         int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2022         int defer_iput = false;
2023
2024         spin_lock(&nilfs->ns_inode_lock);
2025         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2026                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2027                     test_bit(NILFS_I_DIRTY, &ii->i_state))
2028                         continue;
2029
2030                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2031                 brelse(ii->i_bh);
2032                 ii->i_bh = NULL;
2033                 list_del_init(&ii->i_dirty);
2034                 if (!ii->vfs_inode.i_nlink || during_mount) {
2035                         /*
2036                          * Defer calling iput() to avoid deadlocks if
2037                          * i_nlink == 0 or mount is not yet finished.
2038                          */
2039                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2040                         defer_iput = true;
2041                 } else {
2042                         spin_unlock(&nilfs->ns_inode_lock);
2043                         iput(&ii->vfs_inode);
2044                         spin_lock(&nilfs->ns_inode_lock);
2045                 }
2046         }
2047         spin_unlock(&nilfs->ns_inode_lock);
2048
2049         if (defer_iput)
2050                 schedule_work(&sci->sc_iput_work);
2051 }
2052
2053 /*
2054  * Main procedure of segment constructor
2055  */
2056 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2057 {
2058         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2059         int err;
2060
2061         if (sb_rdonly(sci->sc_super))
2062                 return -EROFS;
2063
2064         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2065         sci->sc_cno = nilfs->ns_cno;
2066
2067         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2068         if (unlikely(err))
2069                 goto out;
2070
2071         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2072                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2073
2074         if (nilfs_segctor_clean(sci))
2075                 goto out;
2076
2077         do {
2078                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2079
2080                 err = nilfs_segctor_begin_construction(sci, nilfs);
2081                 if (unlikely(err))
2082                         goto out;
2083
2084                 /* Update time stamp */
2085                 sci->sc_seg_ctime = ktime_get_real_seconds();
2086
2087                 err = nilfs_segctor_collect(sci, nilfs, mode);
2088                 if (unlikely(err))
2089                         goto failed;
2090
2091                 /* Avoid empty segment */
2092                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2093                     nilfs_segbuf_empty(sci->sc_curseg)) {
2094                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2095                         goto out;
2096                 }
2097
2098                 err = nilfs_segctor_assign(sci, mode);
2099                 if (unlikely(err))
2100                         goto failed;
2101
2102                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2103                         nilfs_segctor_fill_in_file_bmap(sci);
2104
2105                 if (mode == SC_LSEG_SR &&
2106                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2107                         err = nilfs_segctor_fill_in_checkpoint(sci);
2108                         if (unlikely(err))
2109                                 goto failed_to_write;
2110
2111                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2112                 }
2113                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2114
2115                 /* Write partial segments */
2116                 nilfs_segctor_prepare_write(sci);
2117
2118                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2119                                             nilfs->ns_crc_seed);
2120
2121                 err = nilfs_segctor_write(sci, nilfs);
2122                 if (unlikely(err))
2123                         goto failed_to_write;
2124
2125                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2126                     nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2127                         /*
2128                          * At this point, we avoid double buffering
2129                          * for blocksize < pagesize because page dirty
2130                          * flag is turned off during write and dirty
2131                          * buffers are not properly collected for
2132                          * pages crossing over segments.
2133                          */
2134                         err = nilfs_segctor_wait(sci);
2135                         if (err)
2136                                 goto failed_to_write;
2137                 }
2138         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2139
2140  out:
2141         nilfs_segctor_drop_written_files(sci, nilfs);
2142         return err;
2143
2144  failed_to_write:
2145         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2146                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2147
2148  failed:
2149         if (nilfs_doing_gc())
2150                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2151         nilfs_segctor_abort_construction(sci, nilfs, err);
2152         goto out;
2153 }
2154
2155 /**
2156  * nilfs_segctor_start_timer - set timer of background write
2157  * @sci: nilfs_sc_info
2158  *
2159  * If the timer has already been set, it ignores the new request.
2160  * This function MUST be called within a section locking the segment
2161  * semaphore.
2162  */
2163 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2164 {
2165         spin_lock(&sci->sc_state_lock);
2166         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2167                 if (sci->sc_task) {
2168                         sci->sc_timer.expires = jiffies + sci->sc_interval;
2169                         add_timer(&sci->sc_timer);
2170                 }
2171                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2172         }
2173         spin_unlock(&sci->sc_state_lock);
2174 }
2175
2176 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2177 {
2178         spin_lock(&sci->sc_state_lock);
2179         if (!(sci->sc_flush_request & BIT(bn))) {
2180                 unsigned long prev_req = sci->sc_flush_request;
2181
2182                 sci->sc_flush_request |= BIT(bn);
2183                 if (!prev_req)
2184                         wake_up(&sci->sc_wait_daemon);
2185         }
2186         spin_unlock(&sci->sc_state_lock);
2187 }
2188
2189 /**
2190  * nilfs_flush_segment - trigger a segment construction for resource control
2191  * @sb: super block
2192  * @ino: inode number of the file to be flushed out.
2193  */
2194 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2195 {
2196         struct the_nilfs *nilfs = sb->s_fs_info;
2197         struct nilfs_sc_info *sci = nilfs->ns_writer;
2198
2199         if (!sci || nilfs_doing_construction())
2200                 return;
2201         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2202                                         /* assign bit 0 to data files */
2203 }
2204
2205 struct nilfs_segctor_wait_request {
2206         wait_queue_entry_t      wq;
2207         __u32           seq;
2208         int             err;
2209         atomic_t        done;
2210 };
2211
2212 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2213 {
2214         struct nilfs_segctor_wait_request wait_req;
2215         int err = 0;
2216
2217         init_wait(&wait_req.wq);
2218         wait_req.err = 0;
2219         atomic_set(&wait_req.done, 0);
2220         init_waitqueue_entry(&wait_req.wq, current);
2221
2222         /*
2223          * To prevent a race issue where completion notifications from the
2224          * log writer thread are missed, increment the request sequence count
2225          * "sc_seq_request" and insert a wait queue entry using the current
2226          * sequence number into the "sc_wait_request" queue at the same time
2227          * within the lock section of "sc_state_lock".
2228          */
2229         spin_lock(&sci->sc_state_lock);
2230         wait_req.seq = ++sci->sc_seq_request;
2231         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2232         spin_unlock(&sci->sc_state_lock);
2233
2234         wake_up(&sci->sc_wait_daemon);
2235
2236         for (;;) {
2237                 set_current_state(TASK_INTERRUPTIBLE);
2238
2239                 /*
2240                  * Synchronize only while the log writer thread is alive.
2241                  * Leave flushing out after the log writer thread exits to
2242                  * the cleanup work in nilfs_segctor_destroy().
2243                  */
2244                 if (!sci->sc_task)
2245                         break;
2246
2247                 if (atomic_read(&wait_req.done)) {
2248                         err = wait_req.err;
2249                         break;
2250                 }
2251                 if (!signal_pending(current)) {
2252                         schedule();
2253                         continue;
2254                 }
2255                 err = -ERESTARTSYS;
2256                 break;
2257         }
2258         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2259         return err;
2260 }
2261
2262 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err, bool force)
2263 {
2264         struct nilfs_segctor_wait_request *wrq, *n;
2265         unsigned long flags;
2266
2267         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2268         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2269                 if (!atomic_read(&wrq->done) &&
2270                     (force || nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq))) {
2271                         wrq->err = err;
2272                         atomic_set(&wrq->done, 1);
2273                 }
2274                 if (atomic_read(&wrq->done)) {
2275                         wrq->wq.func(&wrq->wq,
2276                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2277                                      0, NULL);
2278                 }
2279         }
2280         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2281 }
2282
2283 /**
2284  * nilfs_construct_segment - construct a logical segment
2285  * @sb: super block
2286  *
2287  * Return Value: On success, 0 is retured. On errors, one of the following
2288  * negative error code is returned.
2289  *
2290  * %-EROFS - Read only filesystem.
2291  *
2292  * %-EIO - I/O error
2293  *
2294  * %-ENOSPC - No space left on device (only in a panic state).
2295  *
2296  * %-ERESTARTSYS - Interrupted.
2297  *
2298  * %-ENOMEM - Insufficient memory available.
2299  */
2300 int nilfs_construct_segment(struct super_block *sb)
2301 {
2302         struct the_nilfs *nilfs = sb->s_fs_info;
2303         struct nilfs_sc_info *sci = nilfs->ns_writer;
2304         struct nilfs_transaction_info *ti;
2305         int err;
2306
2307         if (sb_rdonly(sb) || unlikely(!sci))
2308                 return -EROFS;
2309
2310         /* A call inside transactions causes a deadlock. */
2311         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2312
2313         err = nilfs_segctor_sync(sci);
2314         return err;
2315 }
2316
2317 /**
2318  * nilfs_construct_dsync_segment - construct a data-only logical segment
2319  * @sb: super block
2320  * @inode: inode whose data blocks should be written out
2321  * @start: start byte offset
2322  * @end: end byte offset (inclusive)
2323  *
2324  * Return Value: On success, 0 is retured. On errors, one of the following
2325  * negative error code is returned.
2326  *
2327  * %-EROFS - Read only filesystem.
2328  *
2329  * %-EIO - I/O error
2330  *
2331  * %-ENOSPC - No space left on device (only in a panic state).
2332  *
2333  * %-ERESTARTSYS - Interrupted.
2334  *
2335  * %-ENOMEM - Insufficient memory available.
2336  */
2337 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2338                                   loff_t start, loff_t end)
2339 {
2340         struct the_nilfs *nilfs = sb->s_fs_info;
2341         struct nilfs_sc_info *sci = nilfs->ns_writer;
2342         struct nilfs_inode_info *ii;
2343         struct nilfs_transaction_info ti;
2344         int err = 0;
2345
2346         if (sb_rdonly(sb) || unlikely(!sci))
2347                 return -EROFS;
2348
2349         nilfs_transaction_lock(sb, &ti, 0);
2350
2351         ii = NILFS_I(inode);
2352         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2353             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2354             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2355             nilfs_discontinued(nilfs)) {
2356                 nilfs_transaction_unlock(sb);
2357                 err = nilfs_segctor_sync(sci);
2358                 return err;
2359         }
2360
2361         spin_lock(&nilfs->ns_inode_lock);
2362         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2363             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2364                 spin_unlock(&nilfs->ns_inode_lock);
2365                 nilfs_transaction_unlock(sb);
2366                 return 0;
2367         }
2368         spin_unlock(&nilfs->ns_inode_lock);
2369         sci->sc_dsync_inode = ii;
2370         sci->sc_dsync_start = start;
2371         sci->sc_dsync_end = end;
2372
2373         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2374         if (!err)
2375                 nilfs->ns_flushed_device = 0;
2376
2377         nilfs_transaction_unlock(sb);
2378         return err;
2379 }
2380
2381 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2382 #define FLUSH_DAT_BIT   BIT(NILFS_DAT_INO) /* DAT only */
2383
2384 /**
2385  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2386  * @sci: segment constructor object
2387  */
2388 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2389 {
2390         bool thread_is_alive;
2391
2392         spin_lock(&sci->sc_state_lock);
2393         sci->sc_seq_accepted = sci->sc_seq_request;
2394         thread_is_alive = (bool)sci->sc_task;
2395         spin_unlock(&sci->sc_state_lock);
2396
2397         /*
2398          * This function does not race with the log writer thread's
2399          * termination.  Therefore, deleting sc_timer, which should not be
2400          * done after the log writer thread exits, can be done safely outside
2401          * the area protected by sc_state_lock.
2402          */
2403         if (thread_is_alive)
2404                 del_timer_sync(&sci->sc_timer);
2405 }
2406
2407 /**
2408  * nilfs_segctor_notify - notify the result of request to caller threads
2409  * @sci: segment constructor object
2410  * @mode: mode of log forming
2411  * @err: error code to be notified
2412  */
2413 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2414 {
2415         /* Clear requests (even when the construction failed) */
2416         spin_lock(&sci->sc_state_lock);
2417
2418         if (mode == SC_LSEG_SR) {
2419                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2420                 sci->sc_seq_done = sci->sc_seq_accepted;
2421                 nilfs_segctor_wakeup(sci, err, false);
2422                 sci->sc_flush_request = 0;
2423         } else {
2424                 if (mode == SC_FLUSH_FILE)
2425                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2426                 else if (mode == SC_FLUSH_DAT)
2427                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2428
2429                 /* re-enable timer if checkpoint creation was not done */
2430                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) && sci->sc_task &&
2431                     time_before(jiffies, sci->sc_timer.expires))
2432                         add_timer(&sci->sc_timer);
2433         }
2434         spin_unlock(&sci->sc_state_lock);
2435 }
2436
2437 /**
2438  * nilfs_segctor_construct - form logs and write them to disk
2439  * @sci: segment constructor object
2440  * @mode: mode of log forming
2441  */
2442 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2443 {
2444         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2445         struct nilfs_super_block **sbp;
2446         int err = 0;
2447
2448         nilfs_segctor_accept(sci);
2449
2450         if (nilfs_discontinued(nilfs))
2451                 mode = SC_LSEG_SR;
2452         if (!nilfs_segctor_confirm(sci))
2453                 err = nilfs_segctor_do_construct(sci, mode);
2454
2455         if (likely(!err)) {
2456                 if (mode != SC_FLUSH_DAT)
2457                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2458                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2459                     nilfs_discontinued(nilfs)) {
2460                         down_write(&nilfs->ns_sem);
2461                         err = -EIO;
2462                         sbp = nilfs_prepare_super(sci->sc_super,
2463                                                   nilfs_sb_will_flip(nilfs));
2464                         if (likely(sbp)) {
2465                                 nilfs_set_log_cursor(sbp[0], nilfs);
2466                                 err = nilfs_commit_super(sci->sc_super,
2467                                                          NILFS_SB_COMMIT);
2468                         }
2469                         up_write(&nilfs->ns_sem);
2470                 }
2471         }
2472
2473         nilfs_segctor_notify(sci, mode, err);
2474         return err;
2475 }
2476
2477 static void nilfs_construction_timeout(struct timer_list *t)
2478 {
2479         struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2480
2481         wake_up_process(sci->sc_timer_task);
2482 }
2483
2484 static void
2485 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2486 {
2487         struct nilfs_inode_info *ii, *n;
2488
2489         list_for_each_entry_safe(ii, n, head, i_dirty) {
2490                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2491                         continue;
2492                 list_del_init(&ii->i_dirty);
2493                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2494                 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2495                 iput(&ii->vfs_inode);
2496         }
2497 }
2498
2499 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2500                          void **kbufs)
2501 {
2502         struct the_nilfs *nilfs = sb->s_fs_info;
2503         struct nilfs_sc_info *sci = nilfs->ns_writer;
2504         struct nilfs_transaction_info ti;
2505         int err;
2506
2507         if (unlikely(!sci))
2508                 return -EROFS;
2509
2510         nilfs_transaction_lock(sb, &ti, 1);
2511
2512         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2513         if (unlikely(err))
2514                 goto out_unlock;
2515
2516         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2517         if (unlikely(err)) {
2518                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2519                 goto out_unlock;
2520         }
2521
2522         sci->sc_freesegs = kbufs[4];
2523         sci->sc_nfreesegs = argv[4].v_nmembs;
2524         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2525
2526         for (;;) {
2527                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2528                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2529
2530                 if (likely(!err))
2531                         break;
2532
2533                 nilfs_warn(sb, "error %d cleaning segments", err);
2534                 set_current_state(TASK_INTERRUPTIBLE);
2535                 schedule_timeout(sci->sc_interval);
2536         }
2537         if (nilfs_test_opt(nilfs, DISCARD)) {
2538                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2539                                                  sci->sc_nfreesegs);
2540                 if (ret) {
2541                         nilfs_warn(sb,
2542                                    "error %d on discard request, turning discards off for the device",
2543                                    ret);
2544                         nilfs_clear_opt(nilfs, DISCARD);
2545                 }
2546         }
2547
2548  out_unlock:
2549         sci->sc_freesegs = NULL;
2550         sci->sc_nfreesegs = 0;
2551         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2552         nilfs_transaction_unlock(sb);
2553         return err;
2554 }
2555
2556 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2557 {
2558         struct nilfs_transaction_info ti;
2559
2560         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2561         nilfs_segctor_construct(sci, mode);
2562
2563         /*
2564          * Unclosed segment should be retried.  We do this using sc_timer.
2565          * Timeout of sc_timer will invoke complete construction which leads
2566          * to close the current logical segment.
2567          */
2568         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2569                 nilfs_segctor_start_timer(sci);
2570
2571         nilfs_transaction_unlock(sci->sc_super);
2572 }
2573
2574 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2575 {
2576         int mode = 0;
2577
2578         spin_lock(&sci->sc_state_lock);
2579         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2580                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2581         spin_unlock(&sci->sc_state_lock);
2582
2583         if (mode) {
2584                 nilfs_segctor_do_construct(sci, mode);
2585
2586                 spin_lock(&sci->sc_state_lock);
2587                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2588                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2589                 spin_unlock(&sci->sc_state_lock);
2590         }
2591         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2592 }
2593
2594 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2595 {
2596         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2597             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2598                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2599                         return SC_FLUSH_FILE;
2600                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2601                         return SC_FLUSH_DAT;
2602         }
2603         return SC_LSEG_SR;
2604 }
2605
2606 /**
2607  * nilfs_segctor_thread - main loop of the segment constructor thread.
2608  * @arg: pointer to a struct nilfs_sc_info.
2609  *
2610  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2611  * to execute segment constructions.
2612  */
2613 static int nilfs_segctor_thread(void *arg)
2614 {
2615         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2616         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2617         int timeout = 0;
2618
2619         sci->sc_timer_task = current;
2620         timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2621
2622         /* start sync. */
2623         sci->sc_task = current;
2624         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2625         nilfs_info(sci->sc_super,
2626                    "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2627                    sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2628
2629         spin_lock(&sci->sc_state_lock);
2630  loop:
2631         for (;;) {
2632                 int mode;
2633
2634                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2635                         goto end_thread;
2636
2637                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2638                         mode = SC_LSEG_SR;
2639                 else if (sci->sc_flush_request)
2640                         mode = nilfs_segctor_flush_mode(sci);
2641                 else
2642                         break;
2643
2644                 spin_unlock(&sci->sc_state_lock);
2645                 nilfs_segctor_thread_construct(sci, mode);
2646                 spin_lock(&sci->sc_state_lock);
2647                 timeout = 0;
2648         }
2649
2650
2651         if (freezing(current)) {
2652                 spin_unlock(&sci->sc_state_lock);
2653                 try_to_freeze();
2654                 spin_lock(&sci->sc_state_lock);
2655         } else {
2656                 DEFINE_WAIT(wait);
2657                 int should_sleep = 1;
2658
2659                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2660                                 TASK_INTERRUPTIBLE);
2661
2662                 if (sci->sc_seq_request != sci->sc_seq_done)
2663                         should_sleep = 0;
2664                 else if (sci->sc_flush_request)
2665                         should_sleep = 0;
2666                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2667                         should_sleep = time_before(jiffies,
2668                                         sci->sc_timer.expires);
2669
2670                 if (should_sleep) {
2671                         spin_unlock(&sci->sc_state_lock);
2672                         schedule();
2673                         spin_lock(&sci->sc_state_lock);
2674                 }
2675                 finish_wait(&sci->sc_wait_daemon, &wait);
2676                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2677                            time_after_eq(jiffies, sci->sc_timer.expires));
2678
2679                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2680                         set_nilfs_discontinued(nilfs);
2681         }
2682         goto loop;
2683
2684  end_thread:
2685         /* end sync. */
2686         sci->sc_task = NULL;
2687         del_timer_sync(&sci->sc_timer);
2688         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2689         spin_unlock(&sci->sc_state_lock);
2690         return 0;
2691 }
2692
2693 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2694 {
2695         struct task_struct *t;
2696
2697         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2698         if (IS_ERR(t)) {
2699                 int err = PTR_ERR(t);
2700
2701                 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2702                           err);
2703                 return err;
2704         }
2705         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2706         return 0;
2707 }
2708
2709 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2710         __acquires(&sci->sc_state_lock)
2711         __releases(&sci->sc_state_lock)
2712 {
2713         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2714
2715         while (sci->sc_task) {
2716                 wake_up(&sci->sc_wait_daemon);
2717                 spin_unlock(&sci->sc_state_lock);
2718                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2719                 spin_lock(&sci->sc_state_lock);
2720         }
2721 }
2722
2723 /*
2724  * Setup & clean-up functions
2725  */
2726 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2727                                                struct nilfs_root *root)
2728 {
2729         struct the_nilfs *nilfs = sb->s_fs_info;
2730         struct nilfs_sc_info *sci;
2731
2732         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2733         if (!sci)
2734                 return NULL;
2735
2736         sci->sc_super = sb;
2737
2738         nilfs_get_root(root);
2739         sci->sc_root = root;
2740
2741         init_waitqueue_head(&sci->sc_wait_request);
2742         init_waitqueue_head(&sci->sc_wait_daemon);
2743         init_waitqueue_head(&sci->sc_wait_task);
2744         spin_lock_init(&sci->sc_state_lock);
2745         INIT_LIST_HEAD(&sci->sc_dirty_files);
2746         INIT_LIST_HEAD(&sci->sc_segbufs);
2747         INIT_LIST_HEAD(&sci->sc_write_logs);
2748         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2749         INIT_LIST_HEAD(&sci->sc_iput_queue);
2750         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2751
2752         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2753         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2754         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2755
2756         if (nilfs->ns_interval)
2757                 sci->sc_interval = HZ * nilfs->ns_interval;
2758         if (nilfs->ns_watermark)
2759                 sci->sc_watermark = nilfs->ns_watermark;
2760         return sci;
2761 }
2762
2763 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2764 {
2765         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2766
2767         /*
2768          * The segctord thread was stopped and its timer was removed.
2769          * But some tasks remain.
2770          */
2771         do {
2772                 struct nilfs_transaction_info ti;
2773
2774                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2775                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2776                 nilfs_transaction_unlock(sci->sc_super);
2777
2778                 flush_work(&sci->sc_iput_work);
2779
2780         } while (ret && ret != -EROFS && retrycount-- > 0);
2781 }
2782
2783 /**
2784  * nilfs_segctor_destroy - destroy the segment constructor.
2785  * @sci: nilfs_sc_info
2786  *
2787  * nilfs_segctor_destroy() kills the segctord thread and frees
2788  * the nilfs_sc_info struct.
2789  * Caller must hold the segment semaphore.
2790  */
2791 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2792 {
2793         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2794         int flag;
2795
2796         up_write(&nilfs->ns_segctor_sem);
2797
2798         spin_lock(&sci->sc_state_lock);
2799         nilfs_segctor_kill_thread(sci);
2800         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2801                 || sci->sc_seq_request != sci->sc_seq_done);
2802         spin_unlock(&sci->sc_state_lock);
2803
2804         /*
2805          * Forcibly wake up tasks waiting in nilfs_segctor_sync(), which can
2806          * be called from delayed iput() via nilfs_evict_inode() and can race
2807          * with the above log writer thread termination.
2808          */
2809         nilfs_segctor_wakeup(sci, 0, true);
2810
2811         if (flush_work(&sci->sc_iput_work))
2812                 flag = true;
2813
2814         if (flag || !nilfs_segctor_confirm(sci))
2815                 nilfs_segctor_write_out(sci);
2816
2817         if (!list_empty(&sci->sc_dirty_files)) {
2818                 nilfs_warn(sci->sc_super,
2819                            "disposed unprocessed dirty file(s) when stopping log writer");
2820                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2821         }
2822
2823         if (!list_empty(&sci->sc_iput_queue)) {
2824                 nilfs_warn(sci->sc_super,
2825                            "disposed unprocessed inode(s) in iput queue when stopping log writer");
2826                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2827         }
2828
2829         WARN_ON(!list_empty(&sci->sc_segbufs));
2830         WARN_ON(!list_empty(&sci->sc_write_logs));
2831
2832         nilfs_put_root(sci->sc_root);
2833
2834         down_write(&nilfs->ns_segctor_sem);
2835
2836         kfree(sci);
2837 }
2838
2839 /**
2840  * nilfs_attach_log_writer - attach log writer
2841  * @sb: super block instance
2842  * @root: root object of the current filesystem tree
2843  *
2844  * This allocates a log writer object, initializes it, and starts the
2845  * log writer.
2846  *
2847  * Return Value: On success, 0 is returned. On error, one of the following
2848  * negative error code is returned.
2849  *
2850  * %-ENOMEM - Insufficient memory available.
2851  */
2852 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2853 {
2854         struct the_nilfs *nilfs = sb->s_fs_info;
2855         int err;
2856
2857         if (nilfs->ns_writer) {
2858                 /*
2859                  * This happens if the filesystem is made read-only by
2860                  * __nilfs_error or nilfs_remount and then remounted
2861                  * read/write.  In these cases, reuse the existing
2862                  * writer.
2863                  */
2864                 return 0;
2865         }
2866
2867         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2868         if (!nilfs->ns_writer)
2869                 return -ENOMEM;
2870
2871         inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2872
2873         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2874         if (unlikely(err))
2875                 nilfs_detach_log_writer(sb);
2876
2877         return err;
2878 }
2879
2880 /**
2881  * nilfs_detach_log_writer - destroy log writer
2882  * @sb: super block instance
2883  *
2884  * This kills log writer daemon, frees the log writer object, and
2885  * destroys list of dirty files.
2886  */
2887 void nilfs_detach_log_writer(struct super_block *sb)
2888 {
2889         struct the_nilfs *nilfs = sb->s_fs_info;
2890         LIST_HEAD(garbage_list);
2891
2892         down_write(&nilfs->ns_segctor_sem);
2893         if (nilfs->ns_writer) {
2894                 nilfs_segctor_destroy(nilfs->ns_writer);
2895                 nilfs->ns_writer = NULL;
2896         }
2897         set_nilfs_purging(nilfs);
2898
2899         /* Force to free the list of dirty files */
2900         spin_lock(&nilfs->ns_inode_lock);
2901         if (!list_empty(&nilfs->ns_dirty_files)) {
2902                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2903                 nilfs_warn(sb,
2904                            "disposed unprocessed dirty file(s) when detaching log writer");
2905         }
2906         spin_unlock(&nilfs->ns_inode_lock);
2907         up_write(&nilfs->ns_segctor_sem);
2908
2909         nilfs_dispose_list(nilfs, &garbage_list, 1);
2910         clear_nilfs_purging(nilfs);
2911 }