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