fs/nilfs2/segment.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * 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_ge(a, b)   \
 138         (typecheck(__u32, a) && typecheck(__u32, b) && \
 139          ((__s32)(a) - (__s32)(b) >= 0))
 140
 141 static int nilfs_prepare_segment_lock(struct super_block *sb,
 142                                       struct nilfs_transaction_info *ti)
 143 {
 144         struct nilfs_transaction_info *cur_ti = current->journal_info;
 145         void *save = NULL;
 146
 147         if (cur_ti) {
 148                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 149                         return ++cur_ti->ti_count;
 150
 151                 /*
 152                  * If journal_info field is occupied by other FS,
 153                  * it is saved and will be restored on
 154                  * nilfs_transaction_commit().
 155                  */
 156                 nilfs_warn(sb, "journal info from a different FS");
 157                 save = current->journal_info;
 158         }
 159         if (!ti) {
 160                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 161                 if (!ti)
 162                         return -ENOMEM;
 163                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 164         } else {
 165                 ti->ti_flags = 0;
 166         }
 167         ti->ti_count = 0;
 168         ti->ti_save = save;
 169         ti->ti_magic = NILFS_TI_MAGIC;
 170         current->journal_info = ti;
 171         return 0;
 172 }
 173
 174 /**
 175  * nilfs_transaction_begin - start indivisible file operations.
 176  * @sb: super block
 177  * @ti: nilfs_transaction_info
 178  * @vacancy_check: flags for vacancy rate checks
 179  *
 180  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 181  * the segment semaphore, to make a segment construction and write tasks
 182  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 183  * The region enclosed by these two functions can be nested.  To avoid a
 184  * deadlock, the semaphore is only acquired or released in the outermost call.
 185  *
 186  * This function allocates a nilfs_transaction_info struct to keep context
 187  * information on it.  It is initialized and hooked onto the current task in
 188  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 189  * instead; otherwise a new struct is assigned from a slab.
 190  *
 191  * When @vacancy_check flag is set, this function will check the amount of
 192  * free space, and will wait for the GC to reclaim disk space if low capacity.
 193  *
 194  * Return Value: On success, 0 is returned. On error, one of the following
 195  * negative error code is returned.
 196  *
 197  * %-ENOMEM - Insufficient memory available.
 198  *
 199  * %-ENOSPC - No space left on device
 200  */
 201 int nilfs_transaction_begin(struct super_block *sb,
 202                             struct nilfs_transaction_info *ti,
 203                             int vacancy_check)
 204 {
 205         struct the_nilfs *nilfs;
 206         int ret = nilfs_prepare_segment_lock(sb, ti);
 207         struct nilfs_transaction_info *trace_ti;
 208
 209         if (unlikely(ret < 0))
 210                 return ret;
 211         if (ret > 0) {
 212                 trace_ti = current->journal_info;
 213
 214                 trace_nilfs2_transaction_transition(sb, trace_ti,
 215                                     trace_ti->ti_count, trace_ti->ti_flags,
 216                                     TRACE_NILFS2_TRANSACTION_BEGIN);
 217                 return 0;
 218         }
 219
 220         sb_start_intwrite(sb);
 221
 222         nilfs = sb->s_fs_info;
 223         down_read(&nilfs->ns_segctor_sem);
 224         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 225                 up_read(&nilfs->ns_segctor_sem);
 226                 ret = -ENOSPC;
 227                 goto failed;
 228         }
 229
 230         trace_ti = current->journal_info;
 231         trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
 232                                             trace_ti->ti_flags,
 233                                             TRACE_NILFS2_TRANSACTION_BEGIN);
 234         return 0;
 235
 236  failed:
 237         ti = current->journal_info;
 238         current->journal_info = ti->ti_save;
 239         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 240                 kmem_cache_free(nilfs_transaction_cachep, ti);
 241         sb_end_intwrite(sb);
 242         return ret;
 243 }
 244
 245 /**
 246  * nilfs_transaction_commit - commit indivisible file operations.
 247  * @sb: super block
 248  *
 249  * nilfs_transaction_commit() releases the read semaphore which is
 250  * acquired by nilfs_transaction_begin(). This is only performed
 251  * in outermost call of this function.  If a commit flag is set,
 252  * nilfs_transaction_commit() sets a timer to start the segment
 253  * constructor.  If a sync flag is set, it starts construction
 254  * directly.
 255  */
 256 int nilfs_transaction_commit(struct super_block *sb)
 257 {
 258         struct nilfs_transaction_info *ti = current->journal_info;
 259         struct the_nilfs *nilfs = sb->s_fs_info;
 260         int err = 0;
 261
 262         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 263         ti->ti_flags |= NILFS_TI_COMMIT;
 264         if (ti->ti_count > 0) {
 265                 ti->ti_count--;
 266                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 267                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 268                 return 0;
 269         }
 270         if (nilfs->ns_writer) {
 271                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 272
 273                 if (ti->ti_flags & NILFS_TI_COMMIT)
 274                         nilfs_segctor_start_timer(sci);
 275                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 276                         nilfs_segctor_do_flush(sci, 0);
 277         }
 278         up_read(&nilfs->ns_segctor_sem);
 279         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 280                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 281
 282         current->journal_info = ti->ti_save;
 283
 284         if (ti->ti_flags & NILFS_TI_SYNC)
 285                 err = nilfs_construct_segment(sb);
 286         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 287                 kmem_cache_free(nilfs_transaction_cachep, ti);
 288         sb_end_intwrite(sb);
 289         return err;
 290 }
 291
 292 void nilfs_transaction_abort(struct super_block *sb)
 293 {
 294         struct nilfs_transaction_info *ti = current->journal_info;
 295         struct the_nilfs *nilfs = sb->s_fs_info;
 296
 297         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 298         if (ti->ti_count > 0) {
 299                 ti->ti_count--;
 300                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 301                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 302                 return;
 303         }
 304         up_read(&nilfs->ns_segctor_sem);
 305
 306         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 307                     ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 308
 309         current->journal_info = ti->ti_save;
 310         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 311                 kmem_cache_free(nilfs_transaction_cachep, ti);
 312         sb_end_intwrite(sb);
 313 }
 314
 315 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 316 {
 317         struct the_nilfs *nilfs = sb->s_fs_info;
 318         struct nilfs_sc_info *sci = nilfs->ns_writer;
 319
 320         if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request)
 321                 return;
 322
 323         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 324         up_read(&nilfs->ns_segctor_sem);
 325
 326         down_write(&nilfs->ns_segctor_sem);
 327         if (sci->sc_flush_request &&
 328             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 329                 struct nilfs_transaction_info *ti = current->journal_info;
 330
 331                 ti->ti_flags |= NILFS_TI_WRITER;
 332                 nilfs_segctor_do_immediate_flush(sci);
 333                 ti->ti_flags &= ~NILFS_TI_WRITER;
 334         }
 335         downgrade_write(&nilfs->ns_segctor_sem);
 336 }
 337
 338 static void nilfs_transaction_lock(struct super_block *sb,
 339                                    struct nilfs_transaction_info *ti,
 340                                    int gcflag)
 341 {
 342         struct nilfs_transaction_info *cur_ti = current->journal_info;
 343         struct the_nilfs *nilfs = sb->s_fs_info;
 344         struct nilfs_sc_info *sci = nilfs->ns_writer;
 345
 346         WARN_ON(cur_ti);
 347         ti->ti_flags = NILFS_TI_WRITER;
 348         ti->ti_count = 0;
 349         ti->ti_save = cur_ti;
 350         ti->ti_magic = NILFS_TI_MAGIC;
 351         current->journal_info = ti;
 352
 353         for (;;) {
 354                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 355                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
 356
 357                 down_write(&nilfs->ns_segctor_sem);
 358                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 359                         break;
 360
 361                 nilfs_segctor_do_immediate_flush(sci);
 362
 363                 up_write(&nilfs->ns_segctor_sem);
 364                 cond_resched();
 365         }
 366         if (gcflag)
 367                 ti->ti_flags |= NILFS_TI_GC;
 368
 369         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 370                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
 371 }
 372
 373 static void nilfs_transaction_unlock(struct super_block *sb)
 374 {
 375         struct nilfs_transaction_info *ti = current->journal_info;
 376         struct the_nilfs *nilfs = sb->s_fs_info;
 377
 378         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 379         BUG_ON(ti->ti_count > 0);
 380
 381         up_write(&nilfs->ns_segctor_sem);
 382         current->journal_info = ti->ti_save;
 383
 384         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 385                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
 386 }
 387
 388 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 389                                             struct nilfs_segsum_pointer *ssp,
 390                                             unsigned int bytes)
 391 {
 392         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 393         unsigned int blocksize = sci->sc_super->s_blocksize;
 394         void *p;
 395
 396         if (unlikely(ssp->offset + bytes > blocksize)) {
 397                 ssp->offset = 0;
 398                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 399                                                &segbuf->sb_segsum_buffers));
 400                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 401         }
 402         p = ssp->bh->b_data + ssp->offset;
 403         ssp->offset += bytes;
 404         return p;
 405 }
 406
 407 /**
 408  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 409  * @sci: nilfs_sc_info
 410  */
 411 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 412 {
 413         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 414         struct buffer_head *sumbh;
 415         unsigned int sumbytes;
 416         unsigned int flags = 0;
 417         int err;
 418
 419         if (nilfs_doing_gc())
 420                 flags = NILFS_SS_GC;
 421         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 422         if (unlikely(err))
 423                 return err;
 424
 425         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 426         sumbytes = segbuf->sb_sum.sumbytes;
 427         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 428         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 429         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 430         return 0;
 431 }
 432
 433 /**
 434  * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
 435  * @sci: segment constructor object
 436  *
 437  * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
 438  * the current segment summary block.
 439  */
 440 static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
 441 {
 442         struct nilfs_segsum_pointer *ssp;
 443
 444         ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
 445         if (ssp->offset < ssp->bh->b_size)
 446                 memset(ssp->bh->b_data + ssp->offset, 0,
 447                        ssp->bh->b_size - ssp->offset);
 448 }
 449
 450 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 451 {
 452         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 453         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 454                 return -E2BIG; /*
 455                                 * The current segment is filled up
 456                                 * (internal code)
 457                                 */
 458         nilfs_segctor_zeropad_segsum(sci);
 459         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 460         return nilfs_segctor_reset_segment_buffer(sci);
 461 }
 462
 463 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 464 {
 465         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 466         int err;
 467
 468         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 469                 err = nilfs_segctor_feed_segment(sci);
 470                 if (err)
 471                         return err;
 472                 segbuf = sci->sc_curseg;
 473         }
 474         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 475         if (likely(!err))
 476                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 477         return err;
 478 }
 479
 480 /*
 481  * Functions for making segment summary and payloads
 482  */
 483 static int nilfs_segctor_segsum_block_required(
 484         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 485         unsigned int binfo_size)
 486 {
 487         unsigned int blocksize = sci->sc_super->s_blocksize;
 488         /* Size of finfo and binfo is enough small against blocksize */
 489
 490         return ssp->offset + binfo_size +
 491                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 492                 blocksize;
 493 }
 494
 495 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 496                                       struct inode *inode)
 497 {
 498         sci->sc_curseg->sb_sum.nfinfo++;
 499         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 500         nilfs_segctor_map_segsum_entry(
 501                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 502
 503         if (NILFS_I(inode)->i_root &&
 504             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 505                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 506         /* skip finfo */
 507 }
 508
 509 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 510                                     struct inode *inode)
 511 {
 512         struct nilfs_finfo *finfo;
 513         struct nilfs_inode_info *ii;
 514         struct nilfs_segment_buffer *segbuf;
 515         __u64 cno;
 516
 517         if (sci->sc_blk_cnt == 0)
 518                 return;
 519
 520         ii = NILFS_I(inode);
 521
 522         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 523                 cno = ii->i_cno;
 524         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 525                 cno = 0;
 526         else
 527                 cno = sci->sc_cno;
 528
 529         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 530                                                  sizeof(*finfo));
 531         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 532         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 533         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 534         finfo->fi_cno = cpu_to_le64(cno);
 535
 536         segbuf = sci->sc_curseg;
 537         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 538                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 539         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 540         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 541 }
 542
 543 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 544                                         struct buffer_head *bh,
 545                                         struct inode *inode,
 546                                         unsigned int binfo_size)
 547 {
 548         struct nilfs_segment_buffer *segbuf;
 549         int required, err = 0;
 550
 551  retry:
 552         segbuf = sci->sc_curseg;
 553         required = nilfs_segctor_segsum_block_required(
 554                 sci, &sci->sc_binfo_ptr, binfo_size);
 555         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 556                 nilfs_segctor_end_finfo(sci, inode);
 557                 err = nilfs_segctor_feed_segment(sci);
 558                 if (err)
 559                         return err;
 560                 goto retry;
 561         }
 562         if (unlikely(required)) {
 563                 nilfs_segctor_zeropad_segsum(sci);
 564                 err = nilfs_segbuf_extend_segsum(segbuf);
 565                 if (unlikely(err))
 566                         goto failed;
 567         }
 568         if (sci->sc_blk_cnt == 0)
 569                 nilfs_segctor_begin_finfo(sci, inode);
 570
 571         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 572         /* Substitution to vblocknr is delayed until update_blocknr() */
 573         nilfs_segbuf_add_file_buffer(segbuf, bh);
 574         sci->sc_blk_cnt++;
 575  failed:
 576         return err;
 577 }
 578
 579 /*
 580  * Callback functions that enumerate, mark, and collect dirty blocks
 581  */
 582 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 583                                    struct buffer_head *bh, struct inode *inode)
 584 {
 585         int err;
 586
 587         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 588         if (err < 0)
 589                 return err;
 590
 591         err = nilfs_segctor_add_file_block(sci, bh, inode,
 592                                            sizeof(struct nilfs_binfo_v));
 593         if (!err)
 594                 sci->sc_datablk_cnt++;
 595         return err;
 596 }
 597
 598 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 599                                    struct buffer_head *bh,
 600                                    struct inode *inode)
 601 {
 602         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 603 }
 604
 605 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 606                                    struct buffer_head *bh,
 607                                    struct inode *inode)
 608 {
 609         WARN_ON(!buffer_dirty(bh));
 610         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 611 }
 612
 613 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 614                                         struct nilfs_segsum_pointer *ssp,
 615                                         union nilfs_binfo *binfo)
 616 {
 617         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 618                 sci, ssp, sizeof(*binfo_v));
 619         *binfo_v = binfo->bi_v;
 620 }
 621
 622 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 623                                         struct nilfs_segsum_pointer *ssp,
 624                                         union nilfs_binfo *binfo)
 625 {
 626         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 627                 sci, ssp, sizeof(*vblocknr));
 628         *vblocknr = binfo->bi_v.bi_vblocknr;
 629 }
 630
 631 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
 632         .collect_data = nilfs_collect_file_data,
 633         .collect_node = nilfs_collect_file_node,
 634         .collect_bmap = nilfs_collect_file_bmap,
 635         .write_data_binfo = nilfs_write_file_data_binfo,
 636         .write_node_binfo = nilfs_write_file_node_binfo,
 637 };
 638
 639 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 640                                   struct buffer_head *bh, struct inode *inode)
 641 {
 642         int err;
 643
 644         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 645         if (err < 0)
 646                 return err;
 647
 648         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 649         if (!err)
 650                 sci->sc_datablk_cnt++;
 651         return err;
 652 }
 653
 654 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 655                                   struct buffer_head *bh, struct inode *inode)
 656 {
 657         WARN_ON(!buffer_dirty(bh));
 658         return nilfs_segctor_add_file_block(sci, bh, inode,
 659                                             sizeof(struct nilfs_binfo_dat));
 660 }
 661
 662 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 663                                        struct nilfs_segsum_pointer *ssp,
 664                                        union nilfs_binfo *binfo)
 665 {
 666         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 667                                                           sizeof(*blkoff));
 668         *blkoff = binfo->bi_dat.bi_blkoff;
 669 }
 670
 671 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 672                                        struct nilfs_segsum_pointer *ssp,
 673                                        union nilfs_binfo *binfo)
 674 {
 675         struct nilfs_binfo_dat *binfo_dat =
 676                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 677         *binfo_dat = binfo->bi_dat;
 678 }
 679
 680 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
 681         .collect_data = nilfs_collect_dat_data,
 682         .collect_node = nilfs_collect_file_node,
 683         .collect_bmap = nilfs_collect_dat_bmap,
 684         .write_data_binfo = nilfs_write_dat_data_binfo,
 685         .write_node_binfo = nilfs_write_dat_node_binfo,
 686 };
 687
 688 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 689         .collect_data = nilfs_collect_file_data,
 690         .collect_node = NULL,
 691         .collect_bmap = NULL,
 692         .write_data_binfo = nilfs_write_file_data_binfo,
 693         .write_node_binfo = NULL,
 694 };
 695
 696 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 697                                               struct list_head *listp,
 698                                               size_t nlimit,
 699                                               loff_t start, loff_t end)
 700 {
 701         struct address_space *mapping = inode->i_mapping;
 702         struct folio_batch fbatch;
 703         pgoff_t index = 0, last = ULONG_MAX;
 704         size_t ndirties = 0;
 705         int i;
 706
 707         if (unlikely(start != 0 || end != LLONG_MAX)) {
 708                 /*
 709                  * A valid range is given for sync-ing data pages. The
 710                  * range is rounded to per-page; extra dirty buffers
 711                  * may be included if blocksize < pagesize.
 712                  */
 713                 index = start >> PAGE_SHIFT;
 714                 last = end >> PAGE_SHIFT;
 715         }
 716         folio_batch_init(&fbatch);
 717  repeat:
 718         if (unlikely(index > last) ||
 719               !filemap_get_folios_tag(mapping, &index, last,
 720                       PAGECACHE_TAG_DIRTY, &fbatch))
 721                 return ndirties;
 722
 723         for (i = 0; i < folio_batch_count(&fbatch); i++) {
 724                 struct buffer_head *bh, *head;
 725                 struct folio *folio = fbatch.folios[i];
 726
 727                 folio_lock(folio);
 728                 if (unlikely(folio->mapping != mapping)) {
 729                         /* Exclude folios removed from the address space */
 730                         folio_unlock(folio);
 731                         continue;
 732                 }
 733                 head = folio_buffers(folio);
 734                 if (!head)
 735                         head = create_empty_buffers(folio,
 736                                         i_blocksize(inode), 0);
 737                 folio_unlock(folio);
 738
 739                 bh = head;
 740                 do {
 741                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 742                                 continue;
 743                         get_bh(bh);
 744                         list_add_tail(&bh->b_assoc_buffers, listp);
 745                         ndirties++;
 746                         if (unlikely(ndirties >= nlimit)) {
 747                                 folio_batch_release(&fbatch);
 748                                 cond_resched();
 749                                 return ndirties;
 750                         }
 751                 } while (bh = bh->b_this_page, bh != head);
 752         }
 753         folio_batch_release(&fbatch);
 754         cond_resched();
 755         goto repeat;
 756 }
 757
 758 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 759                                             struct list_head *listp)
 760 {
 761         struct nilfs_inode_info *ii = NILFS_I(inode);
 762         struct inode *btnc_inode = ii->i_assoc_inode;
 763         struct folio_batch fbatch;
 764         struct buffer_head *bh, *head;
 765         unsigned int i;
 766         pgoff_t index = 0;
 767
 768         if (!btnc_inode)
 769                 return;
 770         folio_batch_init(&fbatch);
 771
 772         while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
 773                                 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
 774                 for (i = 0; i < folio_batch_count(&fbatch); i++) {
 775                         bh = head = folio_buffers(fbatch.folios[i]);
 776                         do {
 777                                 if (buffer_dirty(bh) &&
 778                                                 !buffer_async_write(bh)) {
 779                                         get_bh(bh);
 780                                         list_add_tail(&bh->b_assoc_buffers,
 781                                                       listp);
 782                                 }
 783                                 bh = bh->b_this_page;
 784                         } while (bh != head);
 785                 }
 786                 folio_batch_release(&fbatch);
 787                 cond_resched();
 788         }
 789 }
 790
 791 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 792                                struct list_head *head, int force)
 793 {
 794         struct nilfs_inode_info *ii, *n;
 795         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 796         unsigned int nv = 0;
 797
 798         while (!list_empty(head)) {
 799                 spin_lock(&nilfs->ns_inode_lock);
 800                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 801                         list_del_init(&ii->i_dirty);
 802                         if (force) {
 803                                 if (unlikely(ii->i_bh)) {
 804                                         brelse(ii->i_bh);
 805                                         ii->i_bh = NULL;
 806                                 }
 807                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 808                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 809                                 list_add_tail(&ii->i_dirty,
 810                                               &nilfs->ns_dirty_files);
 811                                 continue;
 812                         }
 813                         ivec[nv++] = ii;
 814                         if (nv == SC_N_INODEVEC)
 815                                 break;
 816                 }
 817                 spin_unlock(&nilfs->ns_inode_lock);
 818
 819                 for (pii = ivec; nv > 0; pii++, nv--)
 820                         iput(&(*pii)->vfs_inode);
 821         }
 822 }
 823
 824 static void nilfs_iput_work_func(struct work_struct *work)
 825 {
 826         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 827                                                  sc_iput_work);
 828         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 829
 830         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 831 }
 832
 833 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 834                                      struct nilfs_root *root)
 835 {
 836         int ret = 0;
 837
 838         if (nilfs_mdt_fetch_dirty(root->ifile))
 839                 ret++;
 840         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 841                 ret++;
 842         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 843                 ret++;
 844         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 845                 ret++;
 846         return ret;
 847 }
 848
 849 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 850 {
 851         return list_empty(&sci->sc_dirty_files) &&
 852                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 853                 sci->sc_nfreesegs == 0 &&
 854                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 855 }
 856
 857 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 858 {
 859         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 860         int ret = 0;
 861
 862         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 863                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 864
 865         spin_lock(&nilfs->ns_inode_lock);
 866         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 867                 ret++;
 868
 869         spin_unlock(&nilfs->ns_inode_lock);
 870         return ret;
 871 }
 872
 873 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 874 {
 875         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 876
 877         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 878         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 879         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 880         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 881 }
 882
 883 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 884 {
 885         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 886         struct buffer_head *bh_cp;
 887         struct nilfs_checkpoint *raw_cp;
 888         int err;
 889
 890         /* XXX: this interface will be changed */
 891         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 892                                           &raw_cp, &bh_cp);
 893         if (likely(!err)) {
 894                 /*
 895                  * The following code is duplicated with cpfile.  But, it is
 896                  * needed to collect the checkpoint even if it was not newly
 897                  * created.
 898                  */
 899                 mark_buffer_dirty(bh_cp);
 900                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 901                 nilfs_cpfile_put_checkpoint(
 902                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 903         } else if (err == -EINVAL || err == -ENOENT) {
 904                 nilfs_error(sci->sc_super,
 905                             "checkpoint creation failed due to metadata corruption.");
 906                 err = -EIO;
 907         }
 908         return err;
 909 }
 910
 911 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 912 {
 913         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 914         struct buffer_head *bh_cp;
 915         struct nilfs_checkpoint *raw_cp;
 916         int err;
 917
 918         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 919                                           &raw_cp, &bh_cp);
 920         if (unlikely(err)) {
 921                 if (err == -EINVAL || err == -ENOENT) {
 922                         nilfs_error(sci->sc_super,
 923                                     "checkpoint finalization failed due to metadata corruption.");
 924                         err = -EIO;
 925                 }
 926                 goto failed_ibh;
 927         }
 928         raw_cp->cp_snapshot_list.ssl_next = 0;
 929         raw_cp->cp_snapshot_list.ssl_prev = 0;
 930         raw_cp->cp_inodes_count =
 931                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 932         raw_cp->cp_blocks_count =
 933                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 934         raw_cp->cp_nblk_inc =
 935                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 936         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 937         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 938
 939         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 940                 nilfs_checkpoint_clear_minor(raw_cp);
 941         else
 942                 nilfs_checkpoint_set_minor(raw_cp);
 943
 944         nilfs_write_inode_common(sci->sc_root->ifile,
 945                                  &raw_cp->cp_ifile_inode, 1);
 946         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 947         return 0;
 948
 949  failed_ibh:
 950         return err;
 951 }
 952
 953 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 954                                     struct nilfs_inode_info *ii)
 955
 956 {
 957         struct buffer_head *ibh;
 958         struct nilfs_inode *raw_inode;
 959
 960         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 961                 ibh = ii->i_bh;
 962                 BUG_ON(!ibh);
 963                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 964                                                   ibh);
 965                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 966                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 967         }
 968 }
 969
 970 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 971 {
 972         struct nilfs_inode_info *ii;
 973
 974         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 975                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 976                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 977         }
 978 }
 979
 980 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 981                                              struct the_nilfs *nilfs)
 982 {
 983         struct buffer_head *bh_sr;
 984         struct nilfs_super_root *raw_sr;
 985         unsigned int isz, srsz;
 986
 987         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 988
 989         lock_buffer(bh_sr);
 990         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 991         isz = nilfs->ns_inode_size;
 992         srsz = NILFS_SR_BYTES(isz);
 993
 994         raw_sr->sr_sum = 0;  /* Ensure initialization within this update */
 995         raw_sr->sr_bytes = cpu_to_le16(srsz);
 996         raw_sr->sr_nongc_ctime
 997                 = cpu_to_le64(nilfs_doing_gc() ?
 998                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 999         raw_sr->sr_flags = 0;
1000
1001         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
1002                                  NILFS_SR_DAT_OFFSET(isz), 1);
1003         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
1004                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
1005         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
1006                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
1007         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1008         set_buffer_uptodate(bh_sr);
1009         unlock_buffer(bh_sr);
1010 }
1011
1012 static void nilfs_redirty_inodes(struct list_head *head)
1013 {
1014         struct nilfs_inode_info *ii;
1015
1016         list_for_each_entry(ii, head, i_dirty) {
1017                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1018                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1019         }
1020 }
1021
1022 static void nilfs_drop_collected_inodes(struct list_head *head)
1023 {
1024         struct nilfs_inode_info *ii;
1025
1026         list_for_each_entry(ii, head, i_dirty) {
1027                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1028                         continue;
1029
1030                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1031                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1032         }
1033 }
1034
1035 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1036                                        struct inode *inode,
1037                                        struct list_head *listp,
1038                                        int (*collect)(struct nilfs_sc_info *,
1039                                                       struct buffer_head *,
1040                                                       struct inode *))
1041 {
1042         struct buffer_head *bh, *n;
1043         int err = 0;
1044
1045         if (collect) {
1046                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1047                         list_del_init(&bh->b_assoc_buffers);
1048                         err = collect(sci, bh, inode);
1049                         brelse(bh);
1050                         if (unlikely(err))
1051                                 goto dispose_buffers;
1052                 }
1053                 return 0;
1054         }
1055
1056  dispose_buffers:
1057         while (!list_empty(listp)) {
1058                 bh = list_first_entry(listp, struct buffer_head,
1059                                       b_assoc_buffers);
1060                 list_del_init(&bh->b_assoc_buffers);
1061                 brelse(bh);
1062         }
1063         return err;
1064 }
1065
1066 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1067 {
1068         /* Remaining number of blocks within segment buffer */
1069         return sci->sc_segbuf_nblocks -
1070                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1071 }
1072
1073 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1074                                    struct inode *inode,
1075                                    const struct nilfs_sc_operations *sc_ops)
1076 {
1077         LIST_HEAD(data_buffers);
1078         LIST_HEAD(node_buffers);
1079         int err;
1080
1081         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1082                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1083
1084                 n = nilfs_lookup_dirty_data_buffers(
1085                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1086                 if (n > rest) {
1087                         err = nilfs_segctor_apply_buffers(
1088                                 sci, inode, &data_buffers,
1089                                 sc_ops->collect_data);
1090                         BUG_ON(!err); /* always receive -E2BIG or true error */
1091                         goto break_or_fail;
1092                 }
1093         }
1094         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1095
1096         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1097                 err = nilfs_segctor_apply_buffers(
1098                         sci, inode, &data_buffers, sc_ops->collect_data);
1099                 if (unlikely(err)) {
1100                         /* dispose node list */
1101                         nilfs_segctor_apply_buffers(
1102                                 sci, inode, &node_buffers, NULL);
1103                         goto break_or_fail;
1104                 }
1105                 sci->sc_stage.flags |= NILFS_CF_NODE;
1106         }
1107         /* Collect node */
1108         err = nilfs_segctor_apply_buffers(
1109                 sci, inode, &node_buffers, sc_ops->collect_node);
1110         if (unlikely(err))
1111                 goto break_or_fail;
1112
1113         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1114         err = nilfs_segctor_apply_buffers(
1115                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1116         if (unlikely(err))
1117                 goto break_or_fail;
1118
1119         nilfs_segctor_end_finfo(sci, inode);
1120         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1121
1122  break_or_fail:
1123         return err;
1124 }
1125
1126 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1127                                          struct inode *inode)
1128 {
1129         LIST_HEAD(data_buffers);
1130         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1131         int err;
1132
1133         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1134                                             sci->sc_dsync_start,
1135                                             sci->sc_dsync_end);
1136
1137         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1138                                           nilfs_collect_file_data);
1139         if (!err) {
1140                 nilfs_segctor_end_finfo(sci, inode);
1141                 BUG_ON(n > rest);
1142                 /* always receive -E2BIG or true error if n > rest */
1143         }
1144         return err;
1145 }
1146
1147 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1148 {
1149         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1150         struct list_head *head;
1151         struct nilfs_inode_info *ii;
1152         size_t ndone;
1153         int err = 0;
1154
1155         switch (nilfs_sc_cstage_get(sci)) {
1156         case NILFS_ST_INIT:
1157                 /* Pre-processes */
1158                 sci->sc_stage.flags = 0;
1159
1160                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1161                         sci->sc_nblk_inc = 0;
1162                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1163                         if (mode == SC_LSEG_DSYNC) {
1164                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1165                                 goto dsync_mode;
1166                         }
1167                 }
1168
1169                 sci->sc_stage.dirty_file_ptr = NULL;
1170                 sci->sc_stage.gc_inode_ptr = NULL;
1171                 if (mode == SC_FLUSH_DAT) {
1172                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1173                         goto dat_stage;
1174                 }
1175                 nilfs_sc_cstage_inc(sci);
1176                 fallthrough;
1177         case NILFS_ST_GC:
1178                 if (nilfs_doing_gc()) {
1179                         head = &sci->sc_gc_inodes;
1180                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1181                                                 head, i_dirty);
1182                         list_for_each_entry_continue(ii, head, i_dirty) {
1183                                 err = nilfs_segctor_scan_file(
1184                                         sci, &ii->vfs_inode,
1185                                         &nilfs_sc_file_ops);
1186                                 if (unlikely(err)) {
1187                                         sci->sc_stage.gc_inode_ptr = list_entry(
1188                                                 ii->i_dirty.prev,
1189                                                 struct nilfs_inode_info,
1190                                                 i_dirty);
1191                                         goto break_or_fail;
1192                                 }
1193                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1194                         }
1195                         sci->sc_stage.gc_inode_ptr = NULL;
1196                 }
1197                 nilfs_sc_cstage_inc(sci);
1198                 fallthrough;
1199         case NILFS_ST_FILE:
1200                 head = &sci->sc_dirty_files;
1201                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1202                                         i_dirty);
1203                 list_for_each_entry_continue(ii, head, i_dirty) {
1204                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1205
1206                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1207                                                       &nilfs_sc_file_ops);
1208                         if (unlikely(err)) {
1209                                 sci->sc_stage.dirty_file_ptr =
1210                                         list_entry(ii->i_dirty.prev,
1211                                                    struct nilfs_inode_info,
1212                                                    i_dirty);
1213                                 goto break_or_fail;
1214                         }
1215                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1216                         /* XXX: required ? */
1217                 }
1218                 sci->sc_stage.dirty_file_ptr = NULL;
1219                 if (mode == SC_FLUSH_FILE) {
1220                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1221                         return 0;
1222                 }
1223                 nilfs_sc_cstage_inc(sci);
1224                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1225                 fallthrough;
1226         case NILFS_ST_IFILE:
1227                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1228                                               &nilfs_sc_file_ops);
1229                 if (unlikely(err))
1230                         break;
1231                 nilfs_sc_cstage_inc(sci);
1232                 /* Creating a checkpoint */
1233                 err = nilfs_segctor_create_checkpoint(sci);
1234                 if (unlikely(err))
1235                         break;
1236                 fallthrough;
1237         case NILFS_ST_CPFILE:
1238                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1239                                               &nilfs_sc_file_ops);
1240                 if (unlikely(err))
1241                         break;
1242                 nilfs_sc_cstage_inc(sci);
1243                 fallthrough;
1244         case NILFS_ST_SUFILE:
1245                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1246                                          sci->sc_nfreesegs, &ndone);
1247                 if (unlikely(err)) {
1248                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1249                                                   sci->sc_freesegs, ndone,
1250                                                   NULL);
1251                         break;
1252                 }
1253                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1254
1255                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1256                                               &nilfs_sc_file_ops);
1257                 if (unlikely(err))
1258                         break;
1259                 nilfs_sc_cstage_inc(sci);
1260                 fallthrough;
1261         case NILFS_ST_DAT:
1262  dat_stage:
1263                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1264                                               &nilfs_sc_dat_ops);
1265                 if (unlikely(err))
1266                         break;
1267                 if (mode == SC_FLUSH_DAT) {
1268                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1269                         return 0;
1270                 }
1271                 nilfs_sc_cstage_inc(sci);
1272                 fallthrough;
1273         case NILFS_ST_SR:
1274                 if (mode == SC_LSEG_SR) {
1275                         /* Appending a super root */
1276                         err = nilfs_segctor_add_super_root(sci);
1277                         if (unlikely(err))
1278                                 break;
1279                 }
1280                 /* End of a logical segment */
1281                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1282                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1283                 return 0;
1284         case NILFS_ST_DSYNC:
1285  dsync_mode:
1286                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1287                 ii = sci->sc_dsync_inode;
1288                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1289                         break;
1290
1291                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1292                 if (unlikely(err))
1293                         break;
1294                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1295                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1296                 return 0;
1297         case NILFS_ST_DONE:
1298                 return 0;
1299         default:
1300                 BUG();
1301         }
1302
1303  break_or_fail:
1304         return err;
1305 }
1306
1307 /**
1308  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1309  * @sci: nilfs_sc_info
1310  * @nilfs: nilfs object
1311  */
1312 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1313                                             struct the_nilfs *nilfs)
1314 {
1315         struct nilfs_segment_buffer *segbuf, *prev;
1316         __u64 nextnum;
1317         int err, alloc = 0;
1318
1319         segbuf = nilfs_segbuf_new(sci->sc_super);
1320         if (unlikely(!segbuf))
1321                 return -ENOMEM;
1322
1323         if (list_empty(&sci->sc_write_logs)) {
1324                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1325                                  nilfs->ns_pseg_offset, nilfs);
1326                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1327                         nilfs_shift_to_next_segment(nilfs);
1328                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1329                 }
1330
1331                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1332                 nextnum = nilfs->ns_nextnum;
1333
1334                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1335                         /* Start from the head of a new full segment */
1336                         alloc++;
1337         } else {
1338                 /* Continue logs */
1339                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1340                 nilfs_segbuf_map_cont(segbuf, prev);
1341                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1342                 nextnum = prev->sb_nextnum;
1343
1344                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1345                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1346                         segbuf->sb_sum.seg_seq++;
1347                         alloc++;
1348                 }
1349         }
1350
1351         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1352         if (err)
1353                 goto failed;
1354
1355         if (alloc) {
1356                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1357                 if (err)
1358                         goto failed;
1359         }
1360         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1361
1362         BUG_ON(!list_empty(&sci->sc_segbufs));
1363         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1364         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1365         return 0;
1366
1367  failed:
1368         nilfs_segbuf_free(segbuf);
1369         return err;
1370 }
1371
1372 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1373                                          struct the_nilfs *nilfs, int nadd)
1374 {
1375         struct nilfs_segment_buffer *segbuf, *prev;
1376         struct inode *sufile = nilfs->ns_sufile;
1377         __u64 nextnextnum;
1378         LIST_HEAD(list);
1379         int err, ret, i;
1380
1381         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1382         /*
1383          * Since the segment specified with nextnum might be allocated during
1384          * the previous construction, the buffer including its segusage may
1385          * not be dirty.  The following call ensures that the buffer is dirty
1386          * and will pin the buffer on memory until the sufile is written.
1387          */
1388         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1389         if (unlikely(err))
1390                 return err;
1391
1392         for (i = 0; i < nadd; i++) {
1393                 /* extend segment info */
1394                 err = -ENOMEM;
1395                 segbuf = nilfs_segbuf_new(sci->sc_super);
1396                 if (unlikely(!segbuf))
1397                         goto failed;
1398
1399                 /* map this buffer to region of segment on-disk */
1400                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1401                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1402
1403                 /* allocate the next next full segment */
1404                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1405                 if (unlikely(err))
1406                         goto failed_segbuf;
1407
1408                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1409                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1410
1411                 list_add_tail(&segbuf->sb_list, &list);
1412                 prev = segbuf;
1413         }
1414         list_splice_tail(&list, &sci->sc_segbufs);
1415         return 0;
1416
1417  failed_segbuf:
1418         nilfs_segbuf_free(segbuf);
1419  failed:
1420         list_for_each_entry(segbuf, &list, sb_list) {
1421                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1422                 WARN_ON(ret); /* never fails */
1423         }
1424         nilfs_destroy_logs(&list);
1425         return err;
1426 }
1427
1428 static void nilfs_free_incomplete_logs(struct list_head *logs,
1429                                        struct the_nilfs *nilfs)
1430 {
1431         struct nilfs_segment_buffer *segbuf, *prev;
1432         struct inode *sufile = nilfs->ns_sufile;
1433         int ret;
1434
1435         segbuf = NILFS_FIRST_SEGBUF(logs);
1436         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1437                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1438                 WARN_ON(ret); /* never fails */
1439         }
1440         if (atomic_read(&segbuf->sb_err)) {
1441                 /* Case 1: The first segment failed */
1442                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1443                         /*
1444                          * Case 1a:  Partial segment appended into an existing
1445                          * segment
1446                          */
1447                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1448                                                 segbuf->sb_fseg_end);
1449                 else /* Case 1b:  New full segment */
1450                         set_nilfs_discontinued(nilfs);
1451         }
1452
1453         prev = segbuf;
1454         list_for_each_entry_continue(segbuf, logs, sb_list) {
1455                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1456                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1457                         WARN_ON(ret); /* never fails */
1458                 }
1459                 if (atomic_read(&segbuf->sb_err) &&
1460                     segbuf->sb_segnum != nilfs->ns_nextnum)
1461                         /* Case 2: extended segment (!= next) failed */
1462                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1463                 prev = segbuf;
1464         }
1465 }
1466
1467 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1468                                           struct inode *sufile)
1469 {
1470         struct nilfs_segment_buffer *segbuf;
1471         unsigned long live_blocks;
1472         int ret;
1473
1474         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1475                 live_blocks = segbuf->sb_sum.nblocks +
1476                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1477                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1478                                                      live_blocks,
1479                                                      sci->sc_seg_ctime);
1480                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1481         }
1482 }
1483
1484 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1485 {
1486         struct nilfs_segment_buffer *segbuf;
1487         int ret;
1488
1489         segbuf = NILFS_FIRST_SEGBUF(logs);
1490         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1491                                              segbuf->sb_pseg_start -
1492                                              segbuf->sb_fseg_start, 0);
1493         WARN_ON(ret); /* always succeed because the segusage is dirty */
1494
1495         list_for_each_entry_continue(segbuf, logs, sb_list) {
1496                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1497                                                      0, 0);
1498                 WARN_ON(ret); /* always succeed */
1499         }
1500 }
1501
1502 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1503                                             struct nilfs_segment_buffer *last,
1504                                             struct inode *sufile)
1505 {
1506         struct nilfs_segment_buffer *segbuf = last;
1507         int ret;
1508
1509         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1510                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1511                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1512                 WARN_ON(ret);
1513         }
1514         nilfs_truncate_logs(&sci->sc_segbufs, last);
1515 }
1516
1517
1518 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1519                                  struct the_nilfs *nilfs, int mode)
1520 {
1521         struct nilfs_cstage prev_stage = sci->sc_stage;
1522         int err, nadd = 1;
1523
1524         /* Collection retry loop */
1525         for (;;) {
1526                 sci->sc_nblk_this_inc = 0;
1527                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1528
1529                 err = nilfs_segctor_reset_segment_buffer(sci);
1530                 if (unlikely(err))
1531                         goto failed;
1532
1533                 err = nilfs_segctor_collect_blocks(sci, mode);
1534                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1535                 if (!err)
1536                         break;
1537
1538                 if (unlikely(err != -E2BIG))
1539                         goto failed;
1540
1541                 /* The current segment is filled up */
1542                 if (mode != SC_LSEG_SR ||
1543                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1544                         break;
1545
1546                 nilfs_clear_logs(&sci->sc_segbufs);
1547
1548                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1549                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1550                                                         sci->sc_freesegs,
1551                                                         sci->sc_nfreesegs,
1552                                                         NULL);
1553                         WARN_ON(err); /* do not happen */
1554                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1555                 }
1556
1557                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1558                 if (unlikely(err))
1559                         return err;
1560
1561                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1562                 sci->sc_stage = prev_stage;
1563         }
1564         nilfs_segctor_zeropad_segsum(sci);
1565         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1566         return 0;
1567
1568  failed:
1569         return err;
1570 }
1571
1572 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1573                                       struct buffer_head *new_bh)
1574 {
1575         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1576
1577         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1578         /* The caller must release old_bh */
1579 }
1580
1581 static int
1582 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1583                                      struct nilfs_segment_buffer *segbuf,
1584                                      int mode)
1585 {
1586         struct inode *inode = NULL;
1587         sector_t blocknr;
1588         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1589         unsigned long nblocks = 0, ndatablk = 0;
1590         const struct nilfs_sc_operations *sc_op = NULL;
1591         struct nilfs_segsum_pointer ssp;
1592         struct nilfs_finfo *finfo = NULL;
1593         union nilfs_binfo binfo;
1594         struct buffer_head *bh, *bh_org;
1595         ino_t ino = 0;
1596         int err = 0;
1597
1598         if (!nfinfo)
1599                 goto out;
1600
1601         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1602         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1603         ssp.offset = sizeof(struct nilfs_segment_summary);
1604
1605         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1606                 if (bh == segbuf->sb_super_root)
1607                         break;
1608                 if (!finfo) {
1609                         finfo = nilfs_segctor_map_segsum_entry(
1610                                 sci, &ssp, sizeof(*finfo));
1611                         ino = le64_to_cpu(finfo->fi_ino);
1612                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1613                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1614
1615                         inode = bh->b_folio->mapping->host;
1616
1617                         if (mode == SC_LSEG_DSYNC)
1618                                 sc_op = &nilfs_sc_dsync_ops;
1619                         else if (ino == NILFS_DAT_INO)
1620                                 sc_op = &nilfs_sc_dat_ops;
1621                         else /* file blocks */
1622                                 sc_op = &nilfs_sc_file_ops;
1623                 }
1624                 bh_org = bh;
1625                 get_bh(bh_org);
1626                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1627                                         &binfo);
1628                 if (bh != bh_org)
1629                         nilfs_list_replace_buffer(bh_org, bh);
1630                 brelse(bh_org);
1631                 if (unlikely(err))
1632                         goto failed_bmap;
1633
1634                 if (ndatablk > 0)
1635                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1636                 else
1637                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1638
1639                 blocknr++;
1640                 if (--nblocks == 0) {
1641                         finfo = NULL;
1642                         if (--nfinfo == 0)
1643                                 break;
1644                 } else if (ndatablk > 0)
1645                         ndatablk--;
1646         }
1647  out:
1648         return 0;
1649
1650  failed_bmap:
1651         return err;
1652 }
1653
1654 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1655 {
1656         struct nilfs_segment_buffer *segbuf;
1657         int err;
1658
1659         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1660                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1661                 if (unlikely(err))
1662                         return err;
1663                 nilfs_segbuf_fill_in_segsum(segbuf);
1664         }
1665         return 0;
1666 }
1667
1668 static void nilfs_begin_page_io(struct page *page)
1669 {
1670         if (!page || PageWriteback(page))
1671                 /*
1672                  * For split b-tree node pages, this function may be called
1673                  * twice.  We ignore the 2nd or later calls by this check.
1674                  */
1675                 return;
1676
1677         lock_page(page);
1678         clear_page_dirty_for_io(page);
1679         set_page_writeback(page);
1680         unlock_page(page);
1681 }
1682
1683 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1684 {
1685         struct nilfs_segment_buffer *segbuf;
1686         struct page *bd_page = NULL, *fs_page = NULL;
1687
1688         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1689                 struct buffer_head *bh;
1690
1691                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1692                                     b_assoc_buffers) {
1693                         if (bh->b_page != bd_page) {
1694                                 if (bd_page) {
1695                                         lock_page(bd_page);
1696                                         clear_page_dirty_for_io(bd_page);
1697                                         set_page_writeback(bd_page);
1698                                         unlock_page(bd_page);
1699                                 }
1700                                 bd_page = bh->b_page;
1701                         }
1702                 }
1703
1704                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1705                                     b_assoc_buffers) {
1706                         if (bh == segbuf->sb_super_root) {
1707                                 if (bh->b_page != bd_page) {
1708                                         lock_page(bd_page);
1709                                         clear_page_dirty_for_io(bd_page);
1710                                         set_page_writeback(bd_page);
1711                                         unlock_page(bd_page);
1712                                         bd_page = bh->b_page;
1713                                 }
1714                                 break;
1715                         }
1716                         set_buffer_async_write(bh);
1717                         if (bh->b_page != fs_page) {
1718                                 nilfs_begin_page_io(fs_page);
1719                                 fs_page = bh->b_page;
1720                         }
1721                 }
1722         }
1723         if (bd_page) {
1724                 lock_page(bd_page);
1725                 clear_page_dirty_for_io(bd_page);
1726                 set_page_writeback(bd_page);
1727                 unlock_page(bd_page);
1728         }
1729         nilfs_begin_page_io(fs_page);
1730 }
1731
1732 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1733                                struct the_nilfs *nilfs)
1734 {
1735         int ret;
1736
1737         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1738         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1739         return ret;
1740 }
1741
1742 static void nilfs_end_page_io(struct page *page, int err)
1743 {
1744         if (!page)
1745                 return;
1746
1747         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1748                 /*
1749                  * For b-tree node pages, this function may be called twice
1750                  * or more because they might be split in a segment.
1751                  */
1752                 if (PageDirty(page)) {
1753                         /*
1754                          * For pages holding split b-tree node buffers, dirty
1755                          * flag on the buffers may be cleared discretely.
1756                          * In that case, the page is once redirtied for
1757                          * remaining buffers, and it must be cancelled if
1758                          * all the buffers get cleaned later.
1759                          */
1760                         lock_page(page);
1761                         if (nilfs_page_buffers_clean(page))
1762                                 __nilfs_clear_page_dirty(page);
1763                         unlock_page(page);
1764                 }
1765                 return;
1766         }
1767
1768         if (!err) {
1769                 if (!nilfs_page_buffers_clean(page))
1770                         __set_page_dirty_nobuffers(page);
1771                 ClearPageError(page);
1772         } else {
1773                 __set_page_dirty_nobuffers(page);
1774                 SetPageError(page);
1775         }
1776
1777         end_page_writeback(page);
1778 }
1779
1780 static void nilfs_abort_logs(struct list_head *logs, int err)
1781 {
1782         struct nilfs_segment_buffer *segbuf;
1783         struct page *bd_page = NULL, *fs_page = NULL;
1784         struct buffer_head *bh;
1785
1786         if (list_empty(logs))
1787                 return;
1788
1789         list_for_each_entry(segbuf, logs, sb_list) {
1790                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1791                                     b_assoc_buffers) {
1792                         clear_buffer_uptodate(bh);
1793                         if (bh->b_page != bd_page) {
1794                                 if (bd_page)
1795                                         end_page_writeback(bd_page);
1796                                 bd_page = bh->b_page;
1797                         }
1798                 }
1799
1800                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1801                                     b_assoc_buffers) {
1802                         if (bh == segbuf->sb_super_root) {
1803                                 clear_buffer_uptodate(bh);
1804                                 if (bh->b_page != bd_page) {
1805                                         end_page_writeback(bd_page);
1806                                         bd_page = bh->b_page;
1807                                 }
1808                                 break;
1809                         }
1810                         clear_buffer_async_write(bh);
1811                         if (bh->b_page != fs_page) {
1812                                 nilfs_end_page_io(fs_page, err);
1813                                 fs_page = bh->b_page;
1814                         }
1815                 }
1816         }
1817         if (bd_page)
1818                 end_page_writeback(bd_page);
1819
1820         nilfs_end_page_io(fs_page, err);
1821 }
1822
1823 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1824                                              struct the_nilfs *nilfs, int err)
1825 {
1826         LIST_HEAD(logs);
1827         int ret;
1828
1829         list_splice_tail_init(&sci->sc_write_logs, &logs);
1830         ret = nilfs_wait_on_logs(&logs);
1831         nilfs_abort_logs(&logs, ret ? : err);
1832
1833         list_splice_tail_init(&sci->sc_segbufs, &logs);
1834         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1835         nilfs_free_incomplete_logs(&logs, nilfs);
1836
1837         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1838                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1839                                                 sci->sc_freesegs,
1840                                                 sci->sc_nfreesegs,
1841                                                 NULL);
1842                 WARN_ON(ret); /* do not happen */
1843         }
1844
1845         nilfs_destroy_logs(&logs);
1846 }
1847
1848 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1849                                    struct nilfs_segment_buffer *segbuf)
1850 {
1851         nilfs->ns_segnum = segbuf->sb_segnum;
1852         nilfs->ns_nextnum = segbuf->sb_nextnum;
1853         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1854                 + segbuf->sb_sum.nblocks;
1855         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1856         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1857 }
1858
1859 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1860 {
1861         struct nilfs_segment_buffer *segbuf;
1862         struct page *bd_page = NULL, *fs_page = NULL;
1863         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1864         int update_sr = false;
1865
1866         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1867                 struct buffer_head *bh;
1868
1869                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1870                                     b_assoc_buffers) {
1871                         set_buffer_uptodate(bh);
1872                         clear_buffer_dirty(bh);
1873                         if (bh->b_page != bd_page) {
1874                                 if (bd_page)
1875                                         end_page_writeback(bd_page);
1876                                 bd_page = bh->b_page;
1877                         }
1878                 }
1879                 /*
1880                  * We assume that the buffers which belong to the same page
1881                  * continue over the buffer list.
1882                  * Under this assumption, the last BHs of pages is
1883                  * identifiable by the discontinuity of bh->b_page
1884                  * (page != fs_page).
1885                  *
1886                  * For B-tree node blocks, however, this assumption is not
1887                  * guaranteed.  The cleanup code of B-tree node pages needs
1888                  * special care.
1889                  */
1890                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1891                                     b_assoc_buffers) {
1892                         const unsigned long set_bits = BIT(BH_Uptodate);
1893                         const unsigned long clear_bits =
1894                                 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1895                                  BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1896                                  BIT(BH_NILFS_Redirected));
1897
1898                         if (bh == segbuf->sb_super_root) {
1899                                 set_buffer_uptodate(bh);
1900                                 clear_buffer_dirty(bh);
1901                                 if (bh->b_page != bd_page) {
1902                                         end_page_writeback(bd_page);
1903                                         bd_page = bh->b_page;
1904                                 }
1905                                 update_sr = true;
1906                                 break;
1907                         }
1908                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1909                         if (bh->b_page != fs_page) {
1910                                 nilfs_end_page_io(fs_page, 0);
1911                                 fs_page = bh->b_page;
1912                         }
1913                 }
1914
1915                 if (!nilfs_segbuf_simplex(segbuf)) {
1916                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1917                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1918                                 sci->sc_lseg_stime = jiffies;
1919                         }
1920                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1921                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1922                 }
1923         }
1924         /*
1925          * Since pages may continue over multiple segment buffers,
1926          * end of the last page must be checked outside of the loop.
1927          */
1928         if (bd_page)
1929                 end_page_writeback(bd_page);
1930
1931         nilfs_end_page_io(fs_page, 0);
1932
1933         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1934
1935         if (nilfs_doing_gc())
1936                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1937         else
1938                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1939
1940         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1941
1942         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1943         nilfs_set_next_segment(nilfs, segbuf);
1944
1945         if (update_sr) {
1946                 nilfs->ns_flushed_device = 0;
1947                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1948                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1949
1950                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1951                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1952                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1953                 nilfs_segctor_clear_metadata_dirty(sci);
1954         } else
1955                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1956 }
1957
1958 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1959 {
1960         int ret;
1961
1962         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1963         if (!ret) {
1964                 nilfs_segctor_complete_write(sci);
1965                 nilfs_destroy_logs(&sci->sc_write_logs);
1966         }
1967         return ret;
1968 }
1969
1970 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1971                                              struct the_nilfs *nilfs)
1972 {
1973         struct nilfs_inode_info *ii, *n;
1974         struct inode *ifile = sci->sc_root->ifile;
1975
1976         spin_lock(&nilfs->ns_inode_lock);
1977  retry:
1978         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1979                 if (!ii->i_bh) {
1980                         struct buffer_head *ibh;
1981                         int err;
1982
1983                         spin_unlock(&nilfs->ns_inode_lock);
1984                         err = nilfs_ifile_get_inode_block(
1985                                 ifile, ii->vfs_inode.i_ino, &ibh);
1986                         if (unlikely(err)) {
1987                                 nilfs_warn(sci->sc_super,
1988                                            "log writer: error %d getting inode block (ino=%lu)",
1989                                            err, ii->vfs_inode.i_ino);
1990                                 return err;
1991                         }
1992                         spin_lock(&nilfs->ns_inode_lock);
1993                         if (likely(!ii->i_bh))
1994                                 ii->i_bh = ibh;
1995                         else
1996                                 brelse(ibh);
1997                         goto retry;
1998                 }
1999
2000                 // Always redirty the buffer to avoid race condition
2001                 mark_buffer_dirty(ii->i_bh);
2002                 nilfs_mdt_mark_dirty(ifile);
2003
2004                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2005                 set_bit(NILFS_I_BUSY, &ii->i_state);
2006                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
2007         }
2008         spin_unlock(&nilfs->ns_inode_lock);
2009
2010         return 0;
2011 }
2012
2013 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2014                                              struct the_nilfs *nilfs)
2015 {
2016         struct nilfs_inode_info *ii, *n;
2017         int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2018         int defer_iput = false;
2019
2020         spin_lock(&nilfs->ns_inode_lock);
2021         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2022                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2023                     test_bit(NILFS_I_DIRTY, &ii->i_state))
2024                         continue;
2025
2026                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2027                 brelse(ii->i_bh);
2028                 ii->i_bh = NULL;
2029                 list_del_init(&ii->i_dirty);
2030                 if (!ii->vfs_inode.i_nlink || during_mount) {
2031                         /*
2032                          * Defer calling iput() to avoid deadlocks if
2033                          * i_nlink == 0 or mount is not yet finished.
2034                          */
2035                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2036                         defer_iput = true;
2037                 } else {
2038                         spin_unlock(&nilfs->ns_inode_lock);
2039                         iput(&ii->vfs_inode);
2040                         spin_lock(&nilfs->ns_inode_lock);
2041                 }
2042         }
2043         spin_unlock(&nilfs->ns_inode_lock);
2044
2045         if (defer_iput)
2046                 schedule_work(&sci->sc_iput_work);
2047 }
2048
2049 /*
2050  * Main procedure of segment constructor
2051  */
2052 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2053 {
2054         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2055         int err;
2056
2057         if (sb_rdonly(sci->sc_super))
2058                 return -EROFS;
2059
2060         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2061         sci->sc_cno = nilfs->ns_cno;
2062
2063         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2064         if (unlikely(err))
2065                 goto out;
2066
2067         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2068                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2069
2070         if (nilfs_segctor_clean(sci))
2071                 goto out;
2072
2073         do {
2074                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2075
2076                 err = nilfs_segctor_begin_construction(sci, nilfs);
2077                 if (unlikely(err))
2078                         goto out;
2079
2080                 /* Update time stamp */
2081                 sci->sc_seg_ctime = ktime_get_real_seconds();
2082
2083                 err = nilfs_segctor_collect(sci, nilfs, mode);
2084                 if (unlikely(err))
2085                         goto failed;
2086
2087                 /* Avoid empty segment */
2088                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2089                     nilfs_segbuf_empty(sci->sc_curseg)) {
2090                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2091                         goto out;
2092                 }
2093
2094                 err = nilfs_segctor_assign(sci, mode);
2095                 if (unlikely(err))
2096                         goto failed;
2097
2098                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2099                         nilfs_segctor_fill_in_file_bmap(sci);
2100
2101                 if (mode == SC_LSEG_SR &&
2102                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2103                         err = nilfs_segctor_fill_in_checkpoint(sci);
2104                         if (unlikely(err))
2105                                 goto failed_to_write;
2106
2107                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2108                 }
2109                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2110
2111                 /* Write partial segments */
2112                 nilfs_segctor_prepare_write(sci);
2113
2114                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2115                                             nilfs->ns_crc_seed);
2116
2117                 err = nilfs_segctor_write(sci, nilfs);
2118                 if (unlikely(err))
2119                         goto failed_to_write;
2120
2121                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2122                     nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2123                         /*
2124                          * At this point, we avoid double buffering
2125                          * for blocksize < pagesize because page dirty
2126                          * flag is turned off during write and dirty
2127                          * buffers are not properly collected for
2128                          * pages crossing over segments.
2129                          */
2130                         err = nilfs_segctor_wait(sci);
2131                         if (err)
2132                                 goto failed_to_write;
2133                 }
2134         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2135
2136  out:
2137         nilfs_segctor_drop_written_files(sci, nilfs);
2138         return err;
2139
2140  failed_to_write:
2141         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2142                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2143
2144  failed:
2145         if (nilfs_doing_gc())
2146                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2147         nilfs_segctor_abort_construction(sci, nilfs, err);
2148         goto out;
2149 }
2150
2151 /**
2152  * nilfs_segctor_start_timer - set timer of background write
2153  * @sci: nilfs_sc_info
2154  *
2155  * If the timer has already been set, it ignores the new request.
2156  * This function MUST be called within a section locking the segment
2157  * semaphore.
2158  */
2159 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2160 {
2161         spin_lock(&sci->sc_state_lock);
2162         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2163                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2164                 add_timer(&sci->sc_timer);
2165                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2166         }
2167         spin_unlock(&sci->sc_state_lock);
2168 }
2169
2170 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2171 {
2172         spin_lock(&sci->sc_state_lock);
2173         if (!(sci->sc_flush_request & BIT(bn))) {
2174                 unsigned long prev_req = sci->sc_flush_request;
2175
2176                 sci->sc_flush_request |= BIT(bn);
2177                 if (!prev_req)
2178                         wake_up(&sci->sc_wait_daemon);
2179         }
2180         spin_unlock(&sci->sc_state_lock);
2181 }
2182
2183 /**
2184  * nilfs_flush_segment - trigger a segment construction for resource control
2185  * @sb: super block
2186  * @ino: inode number of the file to be flushed out.
2187  */
2188 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2189 {
2190         struct the_nilfs *nilfs = sb->s_fs_info;
2191         struct nilfs_sc_info *sci = nilfs->ns_writer;
2192
2193         if (!sci || nilfs_doing_construction())
2194                 return;
2195         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2196                                         /* assign bit 0 to data files */
2197 }
2198
2199 struct nilfs_segctor_wait_request {
2200         wait_queue_entry_t      wq;
2201         __u32           seq;
2202         int             err;
2203         atomic_t        done;
2204 };
2205
2206 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2207 {
2208         struct nilfs_segctor_wait_request wait_req;
2209         int err = 0;
2210
2211         spin_lock(&sci->sc_state_lock);
2212         init_wait(&wait_req.wq);
2213         wait_req.err = 0;
2214         atomic_set(&wait_req.done, 0);
2215         wait_req.seq = ++sci->sc_seq_request;
2216         spin_unlock(&sci->sc_state_lock);
2217
2218         init_waitqueue_entry(&wait_req.wq, current);
2219         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2220         set_current_state(TASK_INTERRUPTIBLE);
2221         wake_up(&sci->sc_wait_daemon);
2222
2223         for (;;) {
2224                 if (atomic_read(&wait_req.done)) {
2225                         err = wait_req.err;
2226                         break;
2227                 }
2228                 if (!signal_pending(current)) {
2229                         schedule();
2230                         continue;
2231                 }
2232                 err = -ERESTARTSYS;
2233                 break;
2234         }
2235         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2236         return err;
2237 }
2238
2239 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2240 {
2241         struct nilfs_segctor_wait_request *wrq, *n;
2242         unsigned long flags;
2243
2244         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2245         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2246                 if (!atomic_read(&wrq->done) &&
2247                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2248                         wrq->err = err;
2249                         atomic_set(&wrq->done, 1);
2250                 }
2251                 if (atomic_read(&wrq->done)) {
2252                         wrq->wq.func(&wrq->wq,
2253                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2254                                      0, NULL);
2255                 }
2256         }
2257         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2258 }
2259
2260 /**
2261  * nilfs_construct_segment - construct a logical segment
2262  * @sb: super block
2263  *
2264  * Return Value: On success, 0 is returned. On errors, one of the following
2265  * negative error code is returned.
2266  *
2267  * %-EROFS - Read only filesystem.
2268  *
2269  * %-EIO - I/O error
2270  *
2271  * %-ENOSPC - No space left on device (only in a panic state).
2272  *
2273  * %-ERESTARTSYS - Interrupted.
2274  *
2275  * %-ENOMEM - Insufficient memory available.
2276  */
2277 int nilfs_construct_segment(struct super_block *sb)
2278 {
2279         struct the_nilfs *nilfs = sb->s_fs_info;
2280         struct nilfs_sc_info *sci = nilfs->ns_writer;
2281         struct nilfs_transaction_info *ti;
2282
2283         if (sb_rdonly(sb) || unlikely(!sci))
2284                 return -EROFS;
2285
2286         /* A call inside transactions causes a deadlock. */
2287         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2288
2289         return nilfs_segctor_sync(sci);
2290 }
2291
2292 /**
2293  * nilfs_construct_dsync_segment - construct a data-only logical segment
2294  * @sb: super block
2295  * @inode: inode whose data blocks should be written out
2296  * @start: start byte offset
2297  * @end: end byte offset (inclusive)
2298  *
2299  * Return Value: On success, 0 is returned. On errors, one of the following
2300  * negative error code is returned.
2301  *
2302  * %-EROFS - Read only filesystem.
2303  *
2304  * %-EIO - I/O error
2305  *
2306  * %-ENOSPC - No space left on device (only in a panic state).
2307  *
2308  * %-ERESTARTSYS - Interrupted.
2309  *
2310  * %-ENOMEM - Insufficient memory available.
2311  */
2312 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2313                                   loff_t start, loff_t end)
2314 {
2315         struct the_nilfs *nilfs = sb->s_fs_info;
2316         struct nilfs_sc_info *sci = nilfs->ns_writer;
2317         struct nilfs_inode_info *ii;
2318         struct nilfs_transaction_info ti;
2319         int err = 0;
2320
2321         if (sb_rdonly(sb) || unlikely(!sci))
2322                 return -EROFS;
2323
2324         nilfs_transaction_lock(sb, &ti, 0);
2325
2326         ii = NILFS_I(inode);
2327         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2328             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2329             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2330             nilfs_discontinued(nilfs)) {
2331                 nilfs_transaction_unlock(sb);
2332                 err = nilfs_segctor_sync(sci);
2333                 return err;
2334         }
2335
2336         spin_lock(&nilfs->ns_inode_lock);
2337         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2338             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2339                 spin_unlock(&nilfs->ns_inode_lock);
2340                 nilfs_transaction_unlock(sb);
2341                 return 0;
2342         }
2343         spin_unlock(&nilfs->ns_inode_lock);
2344         sci->sc_dsync_inode = ii;
2345         sci->sc_dsync_start = start;
2346         sci->sc_dsync_end = end;
2347
2348         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2349         if (!err)
2350                 nilfs->ns_flushed_device = 0;
2351
2352         nilfs_transaction_unlock(sb);
2353         return err;
2354 }
2355
2356 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2357 #define FLUSH_DAT_BIT   BIT(NILFS_DAT_INO) /* DAT only */
2358
2359 /**
2360  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2361  * @sci: segment constructor object
2362  */
2363 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2364 {
2365         spin_lock(&sci->sc_state_lock);
2366         sci->sc_seq_accepted = sci->sc_seq_request;
2367         spin_unlock(&sci->sc_state_lock);
2368         del_timer_sync(&sci->sc_timer);
2369 }
2370
2371 /**
2372  * nilfs_segctor_notify - notify the result of request to caller threads
2373  * @sci: segment constructor object
2374  * @mode: mode of log forming
2375  * @err: error code to be notified
2376  */
2377 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2378 {
2379         /* Clear requests (even when the construction failed) */
2380         spin_lock(&sci->sc_state_lock);
2381
2382         if (mode == SC_LSEG_SR) {
2383                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2384                 sci->sc_seq_done = sci->sc_seq_accepted;
2385                 nilfs_segctor_wakeup(sci, err);
2386                 sci->sc_flush_request = 0;
2387         } else {
2388                 if (mode == SC_FLUSH_FILE)
2389                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2390                 else if (mode == SC_FLUSH_DAT)
2391                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2392
2393                 /* re-enable timer if checkpoint creation was not done */
2394                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2395                     time_before(jiffies, sci->sc_timer.expires))
2396                         add_timer(&sci->sc_timer);
2397         }
2398         spin_unlock(&sci->sc_state_lock);
2399 }
2400
2401 /**
2402  * nilfs_segctor_construct - form logs and write them to disk
2403  * @sci: segment constructor object
2404  * @mode: mode of log forming
2405  */
2406 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2407 {
2408         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2409         struct nilfs_super_block **sbp;
2410         int err = 0;
2411
2412         nilfs_segctor_accept(sci);
2413
2414         if (nilfs_discontinued(nilfs))
2415                 mode = SC_LSEG_SR;
2416         if (!nilfs_segctor_confirm(sci))
2417                 err = nilfs_segctor_do_construct(sci, mode);
2418
2419         if (likely(!err)) {
2420                 if (mode != SC_FLUSH_DAT)
2421                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2422                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2423                     nilfs_discontinued(nilfs)) {
2424                         down_write(&nilfs->ns_sem);
2425                         err = -EIO;
2426                         sbp = nilfs_prepare_super(sci->sc_super,
2427                                                   nilfs_sb_will_flip(nilfs));
2428                         if (likely(sbp)) {
2429                                 nilfs_set_log_cursor(sbp[0], nilfs);
2430                                 err = nilfs_commit_super(sci->sc_super,
2431                                                          NILFS_SB_COMMIT);
2432                         }
2433                         up_write(&nilfs->ns_sem);
2434                 }
2435         }
2436
2437         nilfs_segctor_notify(sci, mode, err);
2438         return err;
2439 }
2440
2441 static void nilfs_construction_timeout(struct timer_list *t)
2442 {
2443         struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2444
2445         wake_up_process(sci->sc_timer_task);
2446 }
2447
2448 static void
2449 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2450 {
2451         struct nilfs_inode_info *ii, *n;
2452
2453         list_for_each_entry_safe(ii, n, head, i_dirty) {
2454                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2455                         continue;
2456                 list_del_init(&ii->i_dirty);
2457                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2458                 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2459                 iput(&ii->vfs_inode);
2460         }
2461 }
2462
2463 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2464                          void **kbufs)
2465 {
2466         struct the_nilfs *nilfs = sb->s_fs_info;
2467         struct nilfs_sc_info *sci = nilfs->ns_writer;
2468         struct nilfs_transaction_info ti;
2469         int err;
2470
2471         if (unlikely(!sci))
2472                 return -EROFS;
2473
2474         nilfs_transaction_lock(sb, &ti, 1);
2475
2476         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2477         if (unlikely(err))
2478                 goto out_unlock;
2479
2480         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2481         if (unlikely(err)) {
2482                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2483                 goto out_unlock;
2484         }
2485
2486         sci->sc_freesegs = kbufs[4];
2487         sci->sc_nfreesegs = argv[4].v_nmembs;
2488         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2489
2490         for (;;) {
2491                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2492                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2493
2494                 if (likely(!err))
2495                         break;
2496
2497                 nilfs_warn(sb, "error %d cleaning segments", err);
2498                 set_current_state(TASK_INTERRUPTIBLE);
2499                 schedule_timeout(sci->sc_interval);
2500         }
2501         if (nilfs_test_opt(nilfs, DISCARD)) {
2502                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2503                                                  sci->sc_nfreesegs);
2504                 if (ret) {
2505                         nilfs_warn(sb,
2506                                    "error %d on discard request, turning discards off for the device",
2507                                    ret);
2508                         nilfs_clear_opt(nilfs, DISCARD);
2509                 }
2510         }
2511
2512  out_unlock:
2513         sci->sc_freesegs = NULL;
2514         sci->sc_nfreesegs = 0;
2515         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2516         nilfs_transaction_unlock(sb);
2517         return err;
2518 }
2519
2520 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2521 {
2522         struct nilfs_transaction_info ti;
2523
2524         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2525         nilfs_segctor_construct(sci, mode);
2526
2527         /*
2528          * Unclosed segment should be retried.  We do this using sc_timer.
2529          * Timeout of sc_timer will invoke complete construction which leads
2530          * to close the current logical segment.
2531          */
2532         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2533                 nilfs_segctor_start_timer(sci);
2534
2535         nilfs_transaction_unlock(sci->sc_super);
2536 }
2537
2538 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2539 {
2540         int mode = 0;
2541
2542         spin_lock(&sci->sc_state_lock);
2543         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2544                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2545         spin_unlock(&sci->sc_state_lock);
2546
2547         if (mode) {
2548                 nilfs_segctor_do_construct(sci, mode);
2549
2550                 spin_lock(&sci->sc_state_lock);
2551                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2552                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2553                 spin_unlock(&sci->sc_state_lock);
2554         }
2555         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2556 }
2557
2558 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2559 {
2560         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2561             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2562                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2563                         return SC_FLUSH_FILE;
2564                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2565                         return SC_FLUSH_DAT;
2566         }
2567         return SC_LSEG_SR;
2568 }
2569
2570 /**
2571  * nilfs_segctor_thread - main loop of the segment constructor thread.
2572  * @arg: pointer to a struct nilfs_sc_info.
2573  *
2574  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2575  * to execute segment constructions.
2576  */
2577 static int nilfs_segctor_thread(void *arg)
2578 {
2579         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2580         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2581         int timeout = 0;
2582
2583         sci->sc_timer_task = current;
2584
2585         /* start sync. */
2586         sci->sc_task = current;
2587         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2588         nilfs_info(sci->sc_super,
2589                    "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2590                    sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2591
2592         spin_lock(&sci->sc_state_lock);
2593  loop:
2594         for (;;) {
2595                 int mode;
2596
2597                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2598                         goto end_thread;
2599
2600                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2601                         mode = SC_LSEG_SR;
2602                 else if (sci->sc_flush_request)
2603                         mode = nilfs_segctor_flush_mode(sci);
2604                 else
2605                         break;
2606
2607                 spin_unlock(&sci->sc_state_lock);
2608                 nilfs_segctor_thread_construct(sci, mode);
2609                 spin_lock(&sci->sc_state_lock);
2610                 timeout = 0;
2611         }
2612
2613
2614         if (freezing(current)) {
2615                 spin_unlock(&sci->sc_state_lock);
2616                 try_to_freeze();
2617                 spin_lock(&sci->sc_state_lock);
2618         } else {
2619                 DEFINE_WAIT(wait);
2620                 int should_sleep = 1;
2621
2622                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2623                                 TASK_INTERRUPTIBLE);
2624
2625                 if (sci->sc_seq_request != sci->sc_seq_done)
2626                         should_sleep = 0;
2627                 else if (sci->sc_flush_request)
2628                         should_sleep = 0;
2629                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2630                         should_sleep = time_before(jiffies,
2631                                         sci->sc_timer.expires);
2632
2633                 if (should_sleep) {
2634                         spin_unlock(&sci->sc_state_lock);
2635                         schedule();
2636                         spin_lock(&sci->sc_state_lock);
2637                 }
2638                 finish_wait(&sci->sc_wait_daemon, &wait);
2639                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2640                            time_after_eq(jiffies, sci->sc_timer.expires));
2641
2642                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2643                         set_nilfs_discontinued(nilfs);
2644         }
2645         goto loop;
2646
2647  end_thread:
2648         /* end sync. */
2649         sci->sc_task = NULL;
2650         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2651         spin_unlock(&sci->sc_state_lock);
2652         return 0;
2653 }
2654
2655 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2656 {
2657         struct task_struct *t;
2658
2659         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2660         if (IS_ERR(t)) {
2661                 int err = PTR_ERR(t);
2662
2663                 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2664                           err);
2665                 return err;
2666         }
2667         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2668         return 0;
2669 }
2670
2671 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2672         __acquires(&sci->sc_state_lock)
2673         __releases(&sci->sc_state_lock)
2674 {
2675         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2676
2677         while (sci->sc_task) {
2678                 wake_up(&sci->sc_wait_daemon);
2679                 spin_unlock(&sci->sc_state_lock);
2680                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2681                 spin_lock(&sci->sc_state_lock);
2682         }
2683 }
2684
2685 /*
2686  * Setup & clean-up functions
2687  */
2688 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2689                                                struct nilfs_root *root)
2690 {
2691         struct the_nilfs *nilfs = sb->s_fs_info;
2692         struct nilfs_sc_info *sci;
2693
2694         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2695         if (!sci)
2696                 return NULL;
2697
2698         sci->sc_super = sb;
2699
2700         nilfs_get_root(root);
2701         sci->sc_root = root;
2702
2703         init_waitqueue_head(&sci->sc_wait_request);
2704         init_waitqueue_head(&sci->sc_wait_daemon);
2705         init_waitqueue_head(&sci->sc_wait_task);
2706         spin_lock_init(&sci->sc_state_lock);
2707         INIT_LIST_HEAD(&sci->sc_dirty_files);
2708         INIT_LIST_HEAD(&sci->sc_segbufs);
2709         INIT_LIST_HEAD(&sci->sc_write_logs);
2710         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2711         INIT_LIST_HEAD(&sci->sc_iput_queue);
2712         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2713         timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2714
2715         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2716         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2717         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2718
2719         if (nilfs->ns_interval)
2720                 sci->sc_interval = HZ * nilfs->ns_interval;
2721         if (nilfs->ns_watermark)
2722                 sci->sc_watermark = nilfs->ns_watermark;
2723         return sci;
2724 }
2725
2726 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2727 {
2728         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2729
2730         /*
2731          * The segctord thread was stopped and its timer was removed.
2732          * But some tasks remain.
2733          */
2734         do {
2735                 struct nilfs_transaction_info ti;
2736
2737                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2738                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2739                 nilfs_transaction_unlock(sci->sc_super);
2740
2741                 flush_work(&sci->sc_iput_work);
2742
2743         } while (ret && ret != -EROFS && retrycount-- > 0);
2744 }
2745
2746 /**
2747  * nilfs_segctor_destroy - destroy the segment constructor.
2748  * @sci: nilfs_sc_info
2749  *
2750  * nilfs_segctor_destroy() kills the segctord thread and frees
2751  * the nilfs_sc_info struct.
2752  * Caller must hold the segment semaphore.
2753  */
2754 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2755 {
2756         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2757         int flag;
2758
2759         up_write(&nilfs->ns_segctor_sem);
2760
2761         spin_lock(&sci->sc_state_lock);
2762         nilfs_segctor_kill_thread(sci);
2763         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2764                 || sci->sc_seq_request != sci->sc_seq_done);
2765         spin_unlock(&sci->sc_state_lock);
2766
2767         if (flush_work(&sci->sc_iput_work))
2768                 flag = true;
2769
2770         if (flag || !nilfs_segctor_confirm(sci))
2771                 nilfs_segctor_write_out(sci);
2772
2773         if (!list_empty(&sci->sc_dirty_files)) {
2774                 nilfs_warn(sci->sc_super,
2775                            "disposed unprocessed dirty file(s) when stopping log writer");
2776                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2777         }
2778
2779         if (!list_empty(&sci->sc_iput_queue)) {
2780                 nilfs_warn(sci->sc_super,
2781                            "disposed unprocessed inode(s) in iput queue when stopping log writer");
2782                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2783         }
2784
2785         WARN_ON(!list_empty(&sci->sc_segbufs));
2786         WARN_ON(!list_empty(&sci->sc_write_logs));
2787
2788         nilfs_put_root(sci->sc_root);
2789
2790         down_write(&nilfs->ns_segctor_sem);
2791
2792         timer_shutdown_sync(&sci->sc_timer);
2793         kfree(sci);
2794 }
2795
2796 /**
2797  * nilfs_attach_log_writer - attach log writer
2798  * @sb: super block instance
2799  * @root: root object of the current filesystem tree
2800  *
2801  * This allocates a log writer object, initializes it, and starts the
2802  * log writer.
2803  *
2804  * Return Value: On success, 0 is returned. On error, one of the following
2805  * negative error code is returned.
2806  *
2807  * %-ENOMEM - Insufficient memory available.
2808  */
2809 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2810 {
2811         struct the_nilfs *nilfs = sb->s_fs_info;
2812         int err;
2813
2814         if (nilfs->ns_writer) {
2815                 /*
2816                  * This happens if the filesystem is made read-only by
2817                  * __nilfs_error or nilfs_remount and then remounted
2818                  * read/write.  In these cases, reuse the existing
2819                  * writer.
2820                  */
2821                 return 0;
2822         }
2823
2824         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2825         if (!nilfs->ns_writer)
2826                 return -ENOMEM;
2827
2828         inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2829
2830         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2831         if (unlikely(err))
2832                 nilfs_detach_log_writer(sb);
2833
2834         return err;
2835 }
2836
2837 /**
2838  * nilfs_detach_log_writer - destroy log writer
2839  * @sb: super block instance
2840  *
2841  * This kills log writer daemon, frees the log writer object, and
2842  * destroys list of dirty files.
2843  */
2844 void nilfs_detach_log_writer(struct super_block *sb)
2845 {
2846         struct the_nilfs *nilfs = sb->s_fs_info;
2847         LIST_HEAD(garbage_list);
2848
2849         down_write(&nilfs->ns_segctor_sem);
2850         if (nilfs->ns_writer) {
2851                 nilfs_segctor_destroy(nilfs->ns_writer);
2852                 nilfs->ns_writer = NULL;
2853         }
2854         set_nilfs_purging(nilfs);
2855
2856         /* Force to free the list of dirty files */
2857         spin_lock(&nilfs->ns_inode_lock);
2858         if (!list_empty(&nilfs->ns_dirty_files)) {
2859                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2860                 nilfs_warn(sb,
2861                            "disposed unprocessed dirty file(s) when detaching log writer");
2862         }
2863         spin_unlock(&nilfs->ns_inode_lock);
2864         up_write(&nilfs->ns_segctor_sem);
2865
2866         nilfs_dispose_list(nilfs, &garbage_list, 1);
2867         clear_nilfs_purging(nilfs);
2868 }