fs/nilfs2/segment.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * NILFS segment constructor.
   4  *
   5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Ryusuke Konishi.
   8  *
   9  */
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/buffer_head.h>
  13 #include <linux/writeback.h>
  14 #include <linux/bitops.h>
  15 #include <linux/bio.h>
  16 #include <linux/completion.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/backing-dev.h>
  19 #include <linux/freezer.h>
  20 #include <linux/kthread.h>
  21 #include <linux/crc32.h>
  22 #include <linux/pagevec.h>
  23 #include <linux/slab.h>
  24 #include <linux/sched/signal.h>
  25
  26 #include "nilfs.h"
  27 #include "btnode.h"
  28 #include "page.h"
  29 #include "segment.h"
  30 #include "sufile.h"
  31 #include "cpfile.h"
  32 #include "ifile.h"
  33 #include "segbuf.h"
  34
  35
  36 /*
  37  * Segment constructor
  38  */
  39 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  40
  41 #define SC_MAX_SEGDELTA 64   /*
  42                               * Upper limit of the number of segments
  43                               * appended in collection retry loop
  44                               */
  45
  46 /* Construction mode */
  47 enum {
  48         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  49         SC_LSEG_DSYNC,  /*
  50                          * Flush data blocks of a given file and make
  51                          * a logical segment without a super root.
  52                          */
  53         SC_FLUSH_FILE,  /*
  54                          * Flush data files, leads to segment writes without
  55                          * creating a checkpoint.
  56                          */
  57         SC_FLUSH_DAT,   /*
  58                          * Flush DAT file.  This also creates segments
  59                          * without a checkpoint.
  60                          */
  61 };
  62
  63 /* Stage numbers of dirty block collection */
  64 enum {
  65         NILFS_ST_INIT = 0,
  66         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  67         NILFS_ST_FILE,
  68         NILFS_ST_IFILE,
  69         NILFS_ST_CPFILE,
  70         NILFS_ST_SUFILE,
  71         NILFS_ST_DAT,
  72         NILFS_ST_SR,            /* Super root */
  73         NILFS_ST_DSYNC,         /* Data sync blocks */
  74         NILFS_ST_DONE,
  75 };
  76
  77 #define CREATE_TRACE_POINTS
  78 #include <trace/events/nilfs2.h>
  79
  80 /*
  81  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
  82  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
  83  * the variable must use them because transition of stage count must involve
  84  * trace events (trace_nilfs2_collection_stage_transition).
  85  *
  86  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
  87  * produce tracepoint events. It is provided just for making the intention
  88  * clear.
  89  */
  90 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
  91 {
  92         sci->sc_stage.scnt++;
  93         trace_nilfs2_collection_stage_transition(sci);
  94 }
  95
  96 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
  97 {
  98         sci->sc_stage.scnt = next_scnt;
  99         trace_nilfs2_collection_stage_transition(sci);
 100 }
 101
 102 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
 103 {
 104         return sci->sc_stage.scnt;
 105 }
 106
 107 /* State flags of collection */
 108 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
 109 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
 110 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
 111 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
 112
 113 /* Operations depending on the construction mode and file type */
 114 struct nilfs_sc_operations {
 115         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
 116                             struct inode *);
 117         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
 118                             struct inode *);
 119         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
 120                             struct inode *);
 121         void (*write_data_binfo)(struct nilfs_sc_info *,
 122                                  struct nilfs_segsum_pointer *,
 123                                  union nilfs_binfo *);
 124         void (*write_node_binfo)(struct nilfs_sc_info *,
 125                                  struct nilfs_segsum_pointer *,
 126                                  union nilfs_binfo *);
 127 };
 128
 129 /*
 130  * Other definitions
 131  */
 132 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 133 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 134 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 135 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
 136
 137 #define nilfs_cnt32_ge(a, b)   \
 138         (typecheck(__u32, a) && typecheck(__u32, b) && \
 139          ((__s32)(a) - (__s32)(b) >= 0))
 140
 141 static int nilfs_prepare_segment_lock(struct super_block *sb,
 142                                       struct nilfs_transaction_info *ti)
 143 {
 144         struct nilfs_transaction_info *cur_ti = current->journal_info;
 145         void *save = NULL;
 146
 147         if (cur_ti) {
 148                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 149                         return ++cur_ti->ti_count;
 150
 151                 /*
 152                  * If journal_info field is occupied by other FS,
 153                  * it is saved and will be restored on
 154                  * nilfs_transaction_commit().
 155                  */
 156                 nilfs_warn(sb, "journal info from a different FS");
 157                 save = current->journal_info;
 158         }
 159         if (!ti) {
 160                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 161                 if (!ti)
 162                         return -ENOMEM;
 163                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 164         } else {
 165                 ti->ti_flags = 0;
 166         }
 167         ti->ti_count = 0;
 168         ti->ti_save = save;
 169         ti->ti_magic = NILFS_TI_MAGIC;
 170         current->journal_info = ti;
 171         return 0;
 172 }
 173
 174 /**
 175  * nilfs_transaction_begin - start indivisible file operations.
 176  * @sb: super block
 177  * @ti: nilfs_transaction_info
 178  * @vacancy_check: flags for vacancy rate checks
 179  *
 180  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 181  * the segment semaphore, to make a segment construction and write tasks
 182  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 183  * The region enclosed by these two functions can be nested.  To avoid a
 184  * deadlock, the semaphore is only acquired or released in the outermost call.
 185  *
 186  * This function allocates a nilfs_transaction_info struct to keep context
 187  * information on it.  It is initialized and hooked onto the current task in
 188  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 189  * instead; otherwise a new struct is assigned from a slab.
 190  *
 191  * When @vacancy_check flag is set, this function will check the amount of
 192  * free space, and will wait for the GC to reclaim disk space if low capacity.
 193  *
 194  * Return Value: On success, 0 is returned. On error, one of the following
 195  * negative error code is returned.
 196  *
 197  * %-ENOMEM - Insufficient memory available.
 198  *
 199  * %-ENOSPC - No space left on device
 200  */
 201 int nilfs_transaction_begin(struct super_block *sb,
 202                             struct nilfs_transaction_info *ti,
 203                             int vacancy_check)
 204 {
 205         struct the_nilfs *nilfs;
 206         int ret = nilfs_prepare_segment_lock(sb, ti);
 207         struct nilfs_transaction_info *trace_ti;
 208
 209         if (unlikely(ret < 0))
 210                 return ret;
 211         if (ret > 0) {
 212                 trace_ti = current->journal_info;
 213
 214                 trace_nilfs2_transaction_transition(sb, trace_ti,
 215                                     trace_ti->ti_count, trace_ti->ti_flags,
 216                                     TRACE_NILFS2_TRANSACTION_BEGIN);
 217                 return 0;
 218         }
 219
 220         sb_start_intwrite(sb);
 221
 222         nilfs = sb->s_fs_info;
 223         down_read(&nilfs->ns_segctor_sem);
 224         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 225                 up_read(&nilfs->ns_segctor_sem);
 226                 ret = -ENOSPC;
 227                 goto failed;
 228         }
 229
 230         trace_ti = current->journal_info;
 231         trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
 232                                             trace_ti->ti_flags,
 233                                             TRACE_NILFS2_TRANSACTION_BEGIN);
 234         return 0;
 235
 236  failed:
 237         ti = current->journal_info;
 238         current->journal_info = ti->ti_save;
 239         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 240                 kmem_cache_free(nilfs_transaction_cachep, ti);
 241         sb_end_intwrite(sb);
 242         return ret;
 243 }
 244
 245 /**
 246  * nilfs_transaction_commit - commit indivisible file operations.
 247  * @sb: super block
 248  *
 249  * nilfs_transaction_commit() releases the read semaphore which is
 250  * acquired by nilfs_transaction_begin(). This is only performed
 251  * in outermost call of this function.  If a commit flag is set,
 252  * nilfs_transaction_commit() sets a timer to start the segment
 253  * constructor.  If a sync flag is set, it starts construction
 254  * directly.
 255  */
 256 int nilfs_transaction_commit(struct super_block *sb)
 257 {
 258         struct nilfs_transaction_info *ti = current->journal_info;
 259         struct the_nilfs *nilfs = sb->s_fs_info;
 260         int err = 0;
 261
 262         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 263         ti->ti_flags |= NILFS_TI_COMMIT;
 264         if (ti->ti_count > 0) {
 265                 ti->ti_count--;
 266                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 267                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 268                 return 0;
 269         }
 270         if (nilfs->ns_writer) {
 271                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 272
 273                 if (ti->ti_flags & NILFS_TI_COMMIT)
 274                         nilfs_segctor_start_timer(sci);
 275                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 276                         nilfs_segctor_do_flush(sci, 0);
 277         }
 278         up_read(&nilfs->ns_segctor_sem);
 279         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 280                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 281
 282         current->journal_info = ti->ti_save;
 283
 284         if (ti->ti_flags & NILFS_TI_SYNC)
 285                 err = nilfs_construct_segment(sb);
 286         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 287                 kmem_cache_free(nilfs_transaction_cachep, ti);
 288         sb_end_intwrite(sb);
 289         return err;
 290 }
 291
 292 void nilfs_transaction_abort(struct super_block *sb)
 293 {
 294         struct nilfs_transaction_info *ti = current->journal_info;
 295         struct the_nilfs *nilfs = sb->s_fs_info;
 296
 297         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 298         if (ti->ti_count > 0) {
 299                 ti->ti_count--;
 300                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 301                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 302                 return;
 303         }
 304         up_read(&nilfs->ns_segctor_sem);
 305
 306         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 307                     ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 308
 309         current->journal_info = ti->ti_save;
 310         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 311                 kmem_cache_free(nilfs_transaction_cachep, ti);
 312         sb_end_intwrite(sb);
 313 }
 314
 315 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 316 {
 317         struct the_nilfs *nilfs = sb->s_fs_info;
 318         struct nilfs_sc_info *sci = nilfs->ns_writer;
 319
 320         if (sb_rdonly(sb) || unlikely(!sci) || !sci->sc_flush_request)
 321                 return;
 322
 323         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 324         up_read(&nilfs->ns_segctor_sem);
 325
 326         down_write(&nilfs->ns_segctor_sem);
 327         if (sci->sc_flush_request &&
 328             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 329                 struct nilfs_transaction_info *ti = current->journal_info;
 330
 331                 ti->ti_flags |= NILFS_TI_WRITER;
 332                 nilfs_segctor_do_immediate_flush(sci);
 333                 ti->ti_flags &= ~NILFS_TI_WRITER;
 334         }
 335         downgrade_write(&nilfs->ns_segctor_sem);
 336 }
 337
 338 static void nilfs_transaction_lock(struct super_block *sb,
 339                                    struct nilfs_transaction_info *ti,
 340                                    int gcflag)
 341 {
 342         struct nilfs_transaction_info *cur_ti = current->journal_info;
 343         struct the_nilfs *nilfs = sb->s_fs_info;
 344         struct nilfs_sc_info *sci = nilfs->ns_writer;
 345
 346         WARN_ON(cur_ti);
 347         ti->ti_flags = NILFS_TI_WRITER;
 348         ti->ti_count = 0;
 349         ti->ti_save = cur_ti;
 350         ti->ti_magic = NILFS_TI_MAGIC;
 351         current->journal_info = ti;
 352
 353         for (;;) {
 354                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 355                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
 356
 357                 down_write(&nilfs->ns_segctor_sem);
 358                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 359                         break;
 360
 361                 nilfs_segctor_do_immediate_flush(sci);
 362
 363                 up_write(&nilfs->ns_segctor_sem);
 364                 cond_resched();
 365         }
 366         if (gcflag)
 367                 ti->ti_flags |= NILFS_TI_GC;
 368
 369         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 370                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
 371 }
 372
 373 static void nilfs_transaction_unlock(struct super_block *sb)
 374 {
 375         struct nilfs_transaction_info *ti = current->journal_info;
 376         struct the_nilfs *nilfs = sb->s_fs_info;
 377
 378         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 379         BUG_ON(ti->ti_count > 0);
 380
 381         up_write(&nilfs->ns_segctor_sem);
 382         current->journal_info = ti->ti_save;
 383
 384         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 385                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
 386 }
 387
 388 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 389                                             struct nilfs_segsum_pointer *ssp,
 390                                             unsigned int bytes)
 391 {
 392         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 393         unsigned int blocksize = sci->sc_super->s_blocksize;
 394         void *p;
 395
 396         if (unlikely(ssp->offset + bytes > blocksize)) {
 397                 ssp->offset = 0;
 398                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 399                                                &segbuf->sb_segsum_buffers));
 400                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 401         }
 402         p = ssp->bh->b_data + ssp->offset;
 403         ssp->offset += bytes;
 404         return p;
 405 }
 406
 407 /**
 408  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 409  * @sci: nilfs_sc_info
 410  */
 411 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 412 {
 413         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 414         struct buffer_head *sumbh;
 415         unsigned int sumbytes;
 416         unsigned int flags = 0;
 417         int err;
 418
 419         if (nilfs_doing_gc())
 420                 flags = NILFS_SS_GC;
 421         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 422         if (unlikely(err))
 423                 return err;
 424
 425         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 426         sumbytes = segbuf->sb_sum.sumbytes;
 427         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 428         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 429         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 430         return 0;
 431 }
 432
 433 /**
 434  * nilfs_segctor_zeropad_segsum - zero pad the rest of the segment summary area
 435  * @sci: segment constructor object
 436  *
 437  * nilfs_segctor_zeropad_segsum() zero-fills unallocated space at the end of
 438  * the current segment summary block.
 439  */
 440 static void nilfs_segctor_zeropad_segsum(struct nilfs_sc_info *sci)
 441 {
 442         struct nilfs_segsum_pointer *ssp;
 443
 444         ssp = sci->sc_blk_cnt > 0 ? &sci->sc_binfo_ptr : &sci->sc_finfo_ptr;
 445         if (ssp->offset < ssp->bh->b_size)
 446                 memset(ssp->bh->b_data + ssp->offset, 0,
 447                        ssp->bh->b_size - ssp->offset);
 448 }
 449
 450 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 451 {
 452         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 453         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 454                 return -E2BIG; /*
 455                                 * The current segment is filled up
 456                                 * (internal code)
 457                                 */
 458         nilfs_segctor_zeropad_segsum(sci);
 459         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 460         return nilfs_segctor_reset_segment_buffer(sci);
 461 }
 462
 463 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 464 {
 465         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 466         int err;
 467
 468         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 469                 err = nilfs_segctor_feed_segment(sci);
 470                 if (err)
 471                         return err;
 472                 segbuf = sci->sc_curseg;
 473         }
 474         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 475         if (likely(!err))
 476                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 477         return err;
 478 }
 479
 480 /*
 481  * Functions for making segment summary and payloads
 482  */
 483 static int nilfs_segctor_segsum_block_required(
 484         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 485         unsigned int binfo_size)
 486 {
 487         unsigned int blocksize = sci->sc_super->s_blocksize;
 488         /* Size of finfo and binfo is enough small against blocksize */
 489
 490         return ssp->offset + binfo_size +
 491                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 492                 blocksize;
 493 }
 494
 495 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 496                                       struct inode *inode)
 497 {
 498         sci->sc_curseg->sb_sum.nfinfo++;
 499         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 500         nilfs_segctor_map_segsum_entry(
 501                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 502
 503         if (NILFS_I(inode)->i_root &&
 504             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 505                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 506         /* skip finfo */
 507 }
 508
 509 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 510                                     struct inode *inode)
 511 {
 512         struct nilfs_finfo *finfo;
 513         struct nilfs_inode_info *ii;
 514         struct nilfs_segment_buffer *segbuf;
 515         __u64 cno;
 516
 517         if (sci->sc_blk_cnt == 0)
 518                 return;
 519
 520         ii = NILFS_I(inode);
 521
 522         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 523                 cno = ii->i_cno;
 524         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 525                 cno = 0;
 526         else
 527                 cno = sci->sc_cno;
 528
 529         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 530                                                  sizeof(*finfo));
 531         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 532         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 533         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 534         finfo->fi_cno = cpu_to_le64(cno);
 535
 536         segbuf = sci->sc_curseg;
 537         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 538                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 539         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 540         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 541 }
 542
 543 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 544                                         struct buffer_head *bh,
 545                                         struct inode *inode,
 546                                         unsigned int binfo_size)
 547 {
 548         struct nilfs_segment_buffer *segbuf;
 549         int required, err = 0;
 550
 551  retry:
 552         segbuf = sci->sc_curseg;
 553         required = nilfs_segctor_segsum_block_required(
 554                 sci, &sci->sc_binfo_ptr, binfo_size);
 555         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 556                 nilfs_segctor_end_finfo(sci, inode);
 557                 err = nilfs_segctor_feed_segment(sci);
 558                 if (err)
 559                         return err;
 560                 goto retry;
 561         }
 562         if (unlikely(required)) {
 563                 nilfs_segctor_zeropad_segsum(sci);
 564                 err = nilfs_segbuf_extend_segsum(segbuf);
 565                 if (unlikely(err))
 566                         goto failed;
 567         }
 568         if (sci->sc_blk_cnt == 0)
 569                 nilfs_segctor_begin_finfo(sci, inode);
 570
 571         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 572         /* Substitution to vblocknr is delayed until update_blocknr() */
 573         nilfs_segbuf_add_file_buffer(segbuf, bh);
 574         sci->sc_blk_cnt++;
 575  failed:
 576         return err;
 577 }
 578
 579 /*
 580  * Callback functions that enumerate, mark, and collect dirty blocks
 581  */
 582 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 583                                    struct buffer_head *bh, struct inode *inode)
 584 {
 585         int err;
 586
 587         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 588         if (err < 0)
 589                 return err;
 590
 591         err = nilfs_segctor_add_file_block(sci, bh, inode,
 592                                            sizeof(struct nilfs_binfo_v));
 593         if (!err)
 594                 sci->sc_datablk_cnt++;
 595         return err;
 596 }
 597
 598 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 599                                    struct buffer_head *bh,
 600                                    struct inode *inode)
 601 {
 602         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 603 }
 604
 605 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 606                                    struct buffer_head *bh,
 607                                    struct inode *inode)
 608 {
 609         WARN_ON(!buffer_dirty(bh));
 610         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 611 }
 612
 613 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 614                                         struct nilfs_segsum_pointer *ssp,
 615                                         union nilfs_binfo *binfo)
 616 {
 617         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 618                 sci, ssp, sizeof(*binfo_v));
 619         *binfo_v = binfo->bi_v;
 620 }
 621
 622 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 623                                         struct nilfs_segsum_pointer *ssp,
 624                                         union nilfs_binfo *binfo)
 625 {
 626         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 627                 sci, ssp, sizeof(*vblocknr));
 628         *vblocknr = binfo->bi_v.bi_vblocknr;
 629 }
 630
 631 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
 632         .collect_data = nilfs_collect_file_data,
 633         .collect_node = nilfs_collect_file_node,
 634         .collect_bmap = nilfs_collect_file_bmap,
 635         .write_data_binfo = nilfs_write_file_data_binfo,
 636         .write_node_binfo = nilfs_write_file_node_binfo,
 637 };
 638
 639 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 640                                   struct buffer_head *bh, struct inode *inode)
 641 {
 642         int err;
 643
 644         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 645         if (err < 0)
 646                 return err;
 647
 648         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 649         if (!err)
 650                 sci->sc_datablk_cnt++;
 651         return err;
 652 }
 653
 654 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 655                                   struct buffer_head *bh, struct inode *inode)
 656 {
 657         WARN_ON(!buffer_dirty(bh));
 658         return nilfs_segctor_add_file_block(sci, bh, inode,
 659                                             sizeof(struct nilfs_binfo_dat));
 660 }
 661
 662 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 663                                        struct nilfs_segsum_pointer *ssp,
 664                                        union nilfs_binfo *binfo)
 665 {
 666         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 667                                                           sizeof(*blkoff));
 668         *blkoff = binfo->bi_dat.bi_blkoff;
 669 }
 670
 671 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 672                                        struct nilfs_segsum_pointer *ssp,
 673                                        union nilfs_binfo *binfo)
 674 {
 675         struct nilfs_binfo_dat *binfo_dat =
 676                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 677         *binfo_dat = binfo->bi_dat;
 678 }
 679
 680 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
 681         .collect_data = nilfs_collect_dat_data,
 682         .collect_node = nilfs_collect_file_node,
 683         .collect_bmap = nilfs_collect_dat_bmap,
 684         .write_data_binfo = nilfs_write_dat_data_binfo,
 685         .write_node_binfo = nilfs_write_dat_node_binfo,
 686 };
 687
 688 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 689         .collect_data = nilfs_collect_file_data,
 690         .collect_node = NULL,
 691         .collect_bmap = NULL,
 692         .write_data_binfo = nilfs_write_file_data_binfo,
 693         .write_node_binfo = NULL,
 694 };
 695
 696 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 697                                               struct list_head *listp,
 698                                               size_t nlimit,
 699                                               loff_t start, loff_t end)
 700 {
 701         struct address_space *mapping = inode->i_mapping;
 702         struct folio_batch fbatch;
 703         pgoff_t index = 0, last = ULONG_MAX;
 704         size_t ndirties = 0;
 705         int i;
 706
 707         if (unlikely(start != 0 || end != LLONG_MAX)) {
 708                 /*
 709                  * A valid range is given for sync-ing data pages. The
 710                  * range is rounded to per-page; extra dirty buffers
 711                  * may be included if blocksize < pagesize.
 712                  */
 713                 index = start >> PAGE_SHIFT;
 714                 last = end >> PAGE_SHIFT;
 715         }
 716         folio_batch_init(&fbatch);
 717  repeat:
 718         if (unlikely(index > last) ||
 719               !filemap_get_folios_tag(mapping, &index, last,
 720                       PAGECACHE_TAG_DIRTY, &fbatch))
 721                 return ndirties;
 722
 723         for (i = 0; i < folio_batch_count(&fbatch); i++) {
 724                 struct buffer_head *bh, *head;
 725                 struct folio *folio = fbatch.folios[i];
 726
 727                 folio_lock(folio);
 728                 head = folio_buffers(folio);
 729                 if (!head) {
 730                         create_empty_buffers(&folio->page, i_blocksize(inode), 0);
 731                         head = folio_buffers(folio);
 732                 }
 733                 folio_unlock(folio);
 734
 735                 bh = head;
 736                 do {
 737                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 738                                 continue;
 739                         get_bh(bh);
 740                         list_add_tail(&bh->b_assoc_buffers, listp);
 741                         ndirties++;
 742                         if (unlikely(ndirties >= nlimit)) {
 743                                 folio_batch_release(&fbatch);
 744                                 cond_resched();
 745                                 return ndirties;
 746                         }
 747                 } while (bh = bh->b_this_page, bh != head);
 748         }
 749         folio_batch_release(&fbatch);
 750         cond_resched();
 751         goto repeat;
 752 }
 753
 754 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 755                                             struct list_head *listp)
 756 {
 757         struct nilfs_inode_info *ii = NILFS_I(inode);
 758         struct inode *btnc_inode = ii->i_assoc_inode;
 759         struct folio_batch fbatch;
 760         struct buffer_head *bh, *head;
 761         unsigned int i;
 762         pgoff_t index = 0;
 763
 764         if (!btnc_inode)
 765                 return;
 766         folio_batch_init(&fbatch);
 767
 768         while (filemap_get_folios_tag(btnc_inode->i_mapping, &index,
 769                                 (pgoff_t)-1, PAGECACHE_TAG_DIRTY, &fbatch)) {
 770                 for (i = 0; i < folio_batch_count(&fbatch); i++) {
 771                         bh = head = folio_buffers(fbatch.folios[i]);
 772                         do {
 773                                 if (buffer_dirty(bh) &&
 774                                                 !buffer_async_write(bh)) {
 775                                         get_bh(bh);
 776                                         list_add_tail(&bh->b_assoc_buffers,
 777                                                       listp);
 778                                 }
 779                                 bh = bh->b_this_page;
 780                         } while (bh != head);
 781                 }
 782                 folio_batch_release(&fbatch);
 783                 cond_resched();
 784         }
 785 }
 786
 787 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 788                                struct list_head *head, int force)
 789 {
 790         struct nilfs_inode_info *ii, *n;
 791         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 792         unsigned int nv = 0;
 793
 794         while (!list_empty(head)) {
 795                 spin_lock(&nilfs->ns_inode_lock);
 796                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 797                         list_del_init(&ii->i_dirty);
 798                         if (force) {
 799                                 if (unlikely(ii->i_bh)) {
 800                                         brelse(ii->i_bh);
 801                                         ii->i_bh = NULL;
 802                                 }
 803                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 804                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 805                                 list_add_tail(&ii->i_dirty,
 806                                               &nilfs->ns_dirty_files);
 807                                 continue;
 808                         }
 809                         ivec[nv++] = ii;
 810                         if (nv == SC_N_INODEVEC)
 811                                 break;
 812                 }
 813                 spin_unlock(&nilfs->ns_inode_lock);
 814
 815                 for (pii = ivec; nv > 0; pii++, nv--)
 816                         iput(&(*pii)->vfs_inode);
 817         }
 818 }
 819
 820 static void nilfs_iput_work_func(struct work_struct *work)
 821 {
 822         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 823                                                  sc_iput_work);
 824         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 825
 826         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 827 }
 828
 829 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 830                                      struct nilfs_root *root)
 831 {
 832         int ret = 0;
 833
 834         if (nilfs_mdt_fetch_dirty(root->ifile))
 835                 ret++;
 836         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 837                 ret++;
 838         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 839                 ret++;
 840         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 841                 ret++;
 842         return ret;
 843 }
 844
 845 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 846 {
 847         return list_empty(&sci->sc_dirty_files) &&
 848                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 849                 sci->sc_nfreesegs == 0 &&
 850                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 851 }
 852
 853 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 854 {
 855         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 856         int ret = 0;
 857
 858         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 859                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 860
 861         spin_lock(&nilfs->ns_inode_lock);
 862         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 863                 ret++;
 864
 865         spin_unlock(&nilfs->ns_inode_lock);
 866         return ret;
 867 }
 868
 869 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 870 {
 871         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 872
 873         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 874         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 875         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 876         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 877 }
 878
 879 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 880 {
 881         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 882         struct buffer_head *bh_cp;
 883         struct nilfs_checkpoint *raw_cp;
 884         int err;
 885
 886         /* XXX: this interface will be changed */
 887         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 888                                           &raw_cp, &bh_cp);
 889         if (likely(!err)) {
 890                 /*
 891                  * The following code is duplicated with cpfile.  But, it is
 892                  * needed to collect the checkpoint even if it was not newly
 893                  * created.
 894                  */
 895                 mark_buffer_dirty(bh_cp);
 896                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 897                 nilfs_cpfile_put_checkpoint(
 898                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 899         } else if (err == -EINVAL || err == -ENOENT) {
 900                 nilfs_error(sci->sc_super,
 901                             "checkpoint creation failed due to metadata corruption.");
 902                 err = -EIO;
 903         }
 904         return err;
 905 }
 906
 907 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 908 {
 909         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 910         struct buffer_head *bh_cp;
 911         struct nilfs_checkpoint *raw_cp;
 912         int err;
 913
 914         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 915                                           &raw_cp, &bh_cp);
 916         if (unlikely(err)) {
 917                 if (err == -EINVAL || err == -ENOENT) {
 918                         nilfs_error(sci->sc_super,
 919                                     "checkpoint finalization failed due to metadata corruption.");
 920                         err = -EIO;
 921                 }
 922                 goto failed_ibh;
 923         }
 924         raw_cp->cp_snapshot_list.ssl_next = 0;
 925         raw_cp->cp_snapshot_list.ssl_prev = 0;
 926         raw_cp->cp_inodes_count =
 927                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 928         raw_cp->cp_blocks_count =
 929                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 930         raw_cp->cp_nblk_inc =
 931                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 932         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 933         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 934
 935         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 936                 nilfs_checkpoint_clear_minor(raw_cp);
 937         else
 938                 nilfs_checkpoint_set_minor(raw_cp);
 939
 940         nilfs_write_inode_common(sci->sc_root->ifile,
 941                                  &raw_cp->cp_ifile_inode, 1);
 942         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 943         return 0;
 944
 945  failed_ibh:
 946         return err;
 947 }
 948
 949 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 950                                     struct nilfs_inode_info *ii)
 951
 952 {
 953         struct buffer_head *ibh;
 954         struct nilfs_inode *raw_inode;
 955
 956         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 957                 ibh = ii->i_bh;
 958                 BUG_ON(!ibh);
 959                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 960                                                   ibh);
 961                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 962                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 963         }
 964 }
 965
 966 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 967 {
 968         struct nilfs_inode_info *ii;
 969
 970         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 971                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 972                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 973         }
 974 }
 975
 976 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 977                                              struct the_nilfs *nilfs)
 978 {
 979         struct buffer_head *bh_sr;
 980         struct nilfs_super_root *raw_sr;
 981         unsigned int isz, srsz;
 982
 983         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 984         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 985         isz = nilfs->ns_inode_size;
 986         srsz = NILFS_SR_BYTES(isz);
 987
 988         raw_sr->sr_bytes = cpu_to_le16(srsz);
 989         raw_sr->sr_nongc_ctime
 990                 = cpu_to_le64(nilfs_doing_gc() ?
 991                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 992         raw_sr->sr_flags = 0;
 993
 994         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 995                                  NILFS_SR_DAT_OFFSET(isz), 1);
 996         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 997                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 998         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 999                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
1000         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
1001 }
1002
1003 static void nilfs_redirty_inodes(struct list_head *head)
1004 {
1005         struct nilfs_inode_info *ii;
1006
1007         list_for_each_entry(ii, head, i_dirty) {
1008                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1009                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1010         }
1011 }
1012
1013 static void nilfs_drop_collected_inodes(struct list_head *head)
1014 {
1015         struct nilfs_inode_info *ii;
1016
1017         list_for_each_entry(ii, head, i_dirty) {
1018                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1019                         continue;
1020
1021                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1022                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1023         }
1024 }
1025
1026 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1027                                        struct inode *inode,
1028                                        struct list_head *listp,
1029                                        int (*collect)(struct nilfs_sc_info *,
1030                                                       struct buffer_head *,
1031                                                       struct inode *))
1032 {
1033         struct buffer_head *bh, *n;
1034         int err = 0;
1035
1036         if (collect) {
1037                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1038                         list_del_init(&bh->b_assoc_buffers);
1039                         err = collect(sci, bh, inode);
1040                         brelse(bh);
1041                         if (unlikely(err))
1042                                 goto dispose_buffers;
1043                 }
1044                 return 0;
1045         }
1046
1047  dispose_buffers:
1048         while (!list_empty(listp)) {
1049                 bh = list_first_entry(listp, struct buffer_head,
1050                                       b_assoc_buffers);
1051                 list_del_init(&bh->b_assoc_buffers);
1052                 brelse(bh);
1053         }
1054         return err;
1055 }
1056
1057 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1058 {
1059         /* Remaining number of blocks within segment buffer */
1060         return sci->sc_segbuf_nblocks -
1061                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1062 }
1063
1064 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1065                                    struct inode *inode,
1066                                    const struct nilfs_sc_operations *sc_ops)
1067 {
1068         LIST_HEAD(data_buffers);
1069         LIST_HEAD(node_buffers);
1070         int err;
1071
1072         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1073                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1074
1075                 n = nilfs_lookup_dirty_data_buffers(
1076                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1077                 if (n > rest) {
1078                         err = nilfs_segctor_apply_buffers(
1079                                 sci, inode, &data_buffers,
1080                                 sc_ops->collect_data);
1081                         BUG_ON(!err); /* always receive -E2BIG or true error */
1082                         goto break_or_fail;
1083                 }
1084         }
1085         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1086
1087         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1088                 err = nilfs_segctor_apply_buffers(
1089                         sci, inode, &data_buffers, sc_ops->collect_data);
1090                 if (unlikely(err)) {
1091                         /* dispose node list */
1092                         nilfs_segctor_apply_buffers(
1093                                 sci, inode, &node_buffers, NULL);
1094                         goto break_or_fail;
1095                 }
1096                 sci->sc_stage.flags |= NILFS_CF_NODE;
1097         }
1098         /* Collect node */
1099         err = nilfs_segctor_apply_buffers(
1100                 sci, inode, &node_buffers, sc_ops->collect_node);
1101         if (unlikely(err))
1102                 goto break_or_fail;
1103
1104         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1105         err = nilfs_segctor_apply_buffers(
1106                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1107         if (unlikely(err))
1108                 goto break_or_fail;
1109
1110         nilfs_segctor_end_finfo(sci, inode);
1111         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1112
1113  break_or_fail:
1114         return err;
1115 }
1116
1117 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1118                                          struct inode *inode)
1119 {
1120         LIST_HEAD(data_buffers);
1121         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1122         int err;
1123
1124         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1125                                             sci->sc_dsync_start,
1126                                             sci->sc_dsync_end);
1127
1128         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1129                                           nilfs_collect_file_data);
1130         if (!err) {
1131                 nilfs_segctor_end_finfo(sci, inode);
1132                 BUG_ON(n > rest);
1133                 /* always receive -E2BIG or true error if n > rest */
1134         }
1135         return err;
1136 }
1137
1138 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1139 {
1140         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1141         struct list_head *head;
1142         struct nilfs_inode_info *ii;
1143         size_t ndone;
1144         int err = 0;
1145
1146         switch (nilfs_sc_cstage_get(sci)) {
1147         case NILFS_ST_INIT:
1148                 /* Pre-processes */
1149                 sci->sc_stage.flags = 0;
1150
1151                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1152                         sci->sc_nblk_inc = 0;
1153                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1154                         if (mode == SC_LSEG_DSYNC) {
1155                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1156                                 goto dsync_mode;
1157                         }
1158                 }
1159
1160                 sci->sc_stage.dirty_file_ptr = NULL;
1161                 sci->sc_stage.gc_inode_ptr = NULL;
1162                 if (mode == SC_FLUSH_DAT) {
1163                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1164                         goto dat_stage;
1165                 }
1166                 nilfs_sc_cstage_inc(sci);
1167                 fallthrough;
1168         case NILFS_ST_GC:
1169                 if (nilfs_doing_gc()) {
1170                         head = &sci->sc_gc_inodes;
1171                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1172                                                 head, i_dirty);
1173                         list_for_each_entry_continue(ii, head, i_dirty) {
1174                                 err = nilfs_segctor_scan_file(
1175                                         sci, &ii->vfs_inode,
1176                                         &nilfs_sc_file_ops);
1177                                 if (unlikely(err)) {
1178                                         sci->sc_stage.gc_inode_ptr = list_entry(
1179                                                 ii->i_dirty.prev,
1180                                                 struct nilfs_inode_info,
1181                                                 i_dirty);
1182                                         goto break_or_fail;
1183                                 }
1184                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1185                         }
1186                         sci->sc_stage.gc_inode_ptr = NULL;
1187                 }
1188                 nilfs_sc_cstage_inc(sci);
1189                 fallthrough;
1190         case NILFS_ST_FILE:
1191                 head = &sci->sc_dirty_files;
1192                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1193                                         i_dirty);
1194                 list_for_each_entry_continue(ii, head, i_dirty) {
1195                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1196
1197                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1198                                                       &nilfs_sc_file_ops);
1199                         if (unlikely(err)) {
1200                                 sci->sc_stage.dirty_file_ptr =
1201                                         list_entry(ii->i_dirty.prev,
1202                                                    struct nilfs_inode_info,
1203                                                    i_dirty);
1204                                 goto break_or_fail;
1205                         }
1206                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1207                         /* XXX: required ? */
1208                 }
1209                 sci->sc_stage.dirty_file_ptr = NULL;
1210                 if (mode == SC_FLUSH_FILE) {
1211                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1212                         return 0;
1213                 }
1214                 nilfs_sc_cstage_inc(sci);
1215                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1216                 fallthrough;
1217         case NILFS_ST_IFILE:
1218                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1219                                               &nilfs_sc_file_ops);
1220                 if (unlikely(err))
1221                         break;
1222                 nilfs_sc_cstage_inc(sci);
1223                 /* Creating a checkpoint */
1224                 err = nilfs_segctor_create_checkpoint(sci);
1225                 if (unlikely(err))
1226                         break;
1227                 fallthrough;
1228         case NILFS_ST_CPFILE:
1229                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1230                                               &nilfs_sc_file_ops);
1231                 if (unlikely(err))
1232                         break;
1233                 nilfs_sc_cstage_inc(sci);
1234                 fallthrough;
1235         case NILFS_ST_SUFILE:
1236                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1237                                          sci->sc_nfreesegs, &ndone);
1238                 if (unlikely(err)) {
1239                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1240                                                   sci->sc_freesegs, ndone,
1241                                                   NULL);
1242                         break;
1243                 }
1244                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1245
1246                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1247                                               &nilfs_sc_file_ops);
1248                 if (unlikely(err))
1249                         break;
1250                 nilfs_sc_cstage_inc(sci);
1251                 fallthrough;
1252         case NILFS_ST_DAT:
1253  dat_stage:
1254                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1255                                               &nilfs_sc_dat_ops);
1256                 if (unlikely(err))
1257                         break;
1258                 if (mode == SC_FLUSH_DAT) {
1259                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1260                         return 0;
1261                 }
1262                 nilfs_sc_cstage_inc(sci);
1263                 fallthrough;
1264         case NILFS_ST_SR:
1265                 if (mode == SC_LSEG_SR) {
1266                         /* Appending a super root */
1267                         err = nilfs_segctor_add_super_root(sci);
1268                         if (unlikely(err))
1269                                 break;
1270                 }
1271                 /* End of a logical segment */
1272                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1273                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1274                 return 0;
1275         case NILFS_ST_DSYNC:
1276  dsync_mode:
1277                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1278                 ii = sci->sc_dsync_inode;
1279                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1280                         break;
1281
1282                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1283                 if (unlikely(err))
1284                         break;
1285                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1286                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1287                 return 0;
1288         case NILFS_ST_DONE:
1289                 return 0;
1290         default:
1291                 BUG();
1292         }
1293
1294  break_or_fail:
1295         return err;
1296 }
1297
1298 /**
1299  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1300  * @sci: nilfs_sc_info
1301  * @nilfs: nilfs object
1302  */
1303 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1304                                             struct the_nilfs *nilfs)
1305 {
1306         struct nilfs_segment_buffer *segbuf, *prev;
1307         __u64 nextnum;
1308         int err, alloc = 0;
1309
1310         segbuf = nilfs_segbuf_new(sci->sc_super);
1311         if (unlikely(!segbuf))
1312                 return -ENOMEM;
1313
1314         if (list_empty(&sci->sc_write_logs)) {
1315                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1316                                  nilfs->ns_pseg_offset, nilfs);
1317                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1318                         nilfs_shift_to_next_segment(nilfs);
1319                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1320                 }
1321
1322                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1323                 nextnum = nilfs->ns_nextnum;
1324
1325                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1326                         /* Start from the head of a new full segment */
1327                         alloc++;
1328         } else {
1329                 /* Continue logs */
1330                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1331                 nilfs_segbuf_map_cont(segbuf, prev);
1332                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1333                 nextnum = prev->sb_nextnum;
1334
1335                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1336                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1337                         segbuf->sb_sum.seg_seq++;
1338                         alloc++;
1339                 }
1340         }
1341
1342         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1343         if (err)
1344                 goto failed;
1345
1346         if (alloc) {
1347                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1348                 if (err)
1349                         goto failed;
1350         }
1351         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1352
1353         BUG_ON(!list_empty(&sci->sc_segbufs));
1354         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1355         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1356         return 0;
1357
1358  failed:
1359         nilfs_segbuf_free(segbuf);
1360         return err;
1361 }
1362
1363 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1364                                          struct the_nilfs *nilfs, int nadd)
1365 {
1366         struct nilfs_segment_buffer *segbuf, *prev;
1367         struct inode *sufile = nilfs->ns_sufile;
1368         __u64 nextnextnum;
1369         LIST_HEAD(list);
1370         int err, ret, i;
1371
1372         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1373         /*
1374          * Since the segment specified with nextnum might be allocated during
1375          * the previous construction, the buffer including its segusage may
1376          * not be dirty.  The following call ensures that the buffer is dirty
1377          * and will pin the buffer on memory until the sufile is written.
1378          */
1379         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1380         if (unlikely(err))
1381                 return err;
1382
1383         for (i = 0; i < nadd; i++) {
1384                 /* extend segment info */
1385                 err = -ENOMEM;
1386                 segbuf = nilfs_segbuf_new(sci->sc_super);
1387                 if (unlikely(!segbuf))
1388                         goto failed;
1389
1390                 /* map this buffer to region of segment on-disk */
1391                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1392                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1393
1394                 /* allocate the next next full segment */
1395                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1396                 if (unlikely(err))
1397                         goto failed_segbuf;
1398
1399                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1400                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1401
1402                 list_add_tail(&segbuf->sb_list, &list);
1403                 prev = segbuf;
1404         }
1405         list_splice_tail(&list, &sci->sc_segbufs);
1406         return 0;
1407
1408  failed_segbuf:
1409         nilfs_segbuf_free(segbuf);
1410  failed:
1411         list_for_each_entry(segbuf, &list, sb_list) {
1412                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1413                 WARN_ON(ret); /* never fails */
1414         }
1415         nilfs_destroy_logs(&list);
1416         return err;
1417 }
1418
1419 static void nilfs_free_incomplete_logs(struct list_head *logs,
1420                                        struct the_nilfs *nilfs)
1421 {
1422         struct nilfs_segment_buffer *segbuf, *prev;
1423         struct inode *sufile = nilfs->ns_sufile;
1424         int ret;
1425
1426         segbuf = NILFS_FIRST_SEGBUF(logs);
1427         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1428                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1429                 WARN_ON(ret); /* never fails */
1430         }
1431         if (atomic_read(&segbuf->sb_err)) {
1432                 /* Case 1: The first segment failed */
1433                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1434                         /*
1435                          * Case 1a:  Partial segment appended into an existing
1436                          * segment
1437                          */
1438                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1439                                                 segbuf->sb_fseg_end);
1440                 else /* Case 1b:  New full segment */
1441                         set_nilfs_discontinued(nilfs);
1442         }
1443
1444         prev = segbuf;
1445         list_for_each_entry_continue(segbuf, logs, sb_list) {
1446                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1447                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1448                         WARN_ON(ret); /* never fails */
1449                 }
1450                 if (atomic_read(&segbuf->sb_err) &&
1451                     segbuf->sb_segnum != nilfs->ns_nextnum)
1452                         /* Case 2: extended segment (!= next) failed */
1453                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1454                 prev = segbuf;
1455         }
1456 }
1457
1458 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1459                                           struct inode *sufile)
1460 {
1461         struct nilfs_segment_buffer *segbuf;
1462         unsigned long live_blocks;
1463         int ret;
1464
1465         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1466                 live_blocks = segbuf->sb_sum.nblocks +
1467                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1468                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1469                                                      live_blocks,
1470                                                      sci->sc_seg_ctime);
1471                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1472         }
1473 }
1474
1475 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1476 {
1477         struct nilfs_segment_buffer *segbuf;
1478         int ret;
1479
1480         segbuf = NILFS_FIRST_SEGBUF(logs);
1481         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1482                                              segbuf->sb_pseg_start -
1483                                              segbuf->sb_fseg_start, 0);
1484         WARN_ON(ret); /* always succeed because the segusage is dirty */
1485
1486         list_for_each_entry_continue(segbuf, logs, sb_list) {
1487                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1488                                                      0, 0);
1489                 WARN_ON(ret); /* always succeed */
1490         }
1491 }
1492
1493 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1494                                             struct nilfs_segment_buffer *last,
1495                                             struct inode *sufile)
1496 {
1497         struct nilfs_segment_buffer *segbuf = last;
1498         int ret;
1499
1500         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1501                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1502                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1503                 WARN_ON(ret);
1504         }
1505         nilfs_truncate_logs(&sci->sc_segbufs, last);
1506 }
1507
1508
1509 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1510                                  struct the_nilfs *nilfs, int mode)
1511 {
1512         struct nilfs_cstage prev_stage = sci->sc_stage;
1513         int err, nadd = 1;
1514
1515         /* Collection retry loop */
1516         for (;;) {
1517                 sci->sc_nblk_this_inc = 0;
1518                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1519
1520                 err = nilfs_segctor_reset_segment_buffer(sci);
1521                 if (unlikely(err))
1522                         goto failed;
1523
1524                 err = nilfs_segctor_collect_blocks(sci, mode);
1525                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1526                 if (!err)
1527                         break;
1528
1529                 if (unlikely(err != -E2BIG))
1530                         goto failed;
1531
1532                 /* The current segment is filled up */
1533                 if (mode != SC_LSEG_SR ||
1534                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1535                         break;
1536
1537                 nilfs_clear_logs(&sci->sc_segbufs);
1538
1539                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1540                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1541                                                         sci->sc_freesegs,
1542                                                         sci->sc_nfreesegs,
1543                                                         NULL);
1544                         WARN_ON(err); /* do not happen */
1545                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1546                 }
1547
1548                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1549                 if (unlikely(err))
1550                         return err;
1551
1552                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1553                 sci->sc_stage = prev_stage;
1554         }
1555         nilfs_segctor_zeropad_segsum(sci);
1556         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1557         return 0;
1558
1559  failed:
1560         return err;
1561 }
1562
1563 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1564                                       struct buffer_head *new_bh)
1565 {
1566         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1567
1568         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1569         /* The caller must release old_bh */
1570 }
1571
1572 static int
1573 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1574                                      struct nilfs_segment_buffer *segbuf,
1575                                      int mode)
1576 {
1577         struct inode *inode = NULL;
1578         sector_t blocknr;
1579         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1580         unsigned long nblocks = 0, ndatablk = 0;
1581         const struct nilfs_sc_operations *sc_op = NULL;
1582         struct nilfs_segsum_pointer ssp;
1583         struct nilfs_finfo *finfo = NULL;
1584         union nilfs_binfo binfo;
1585         struct buffer_head *bh, *bh_org;
1586         ino_t ino = 0;
1587         int err = 0;
1588
1589         if (!nfinfo)
1590                 goto out;
1591
1592         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1593         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1594         ssp.offset = sizeof(struct nilfs_segment_summary);
1595
1596         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1597                 if (bh == segbuf->sb_super_root)
1598                         break;
1599                 if (!finfo) {
1600                         finfo = nilfs_segctor_map_segsum_entry(
1601                                 sci, &ssp, sizeof(*finfo));
1602                         ino = le64_to_cpu(finfo->fi_ino);
1603                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1604                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1605
1606                         inode = bh->b_folio->mapping->host;
1607
1608                         if (mode == SC_LSEG_DSYNC)
1609                                 sc_op = &nilfs_sc_dsync_ops;
1610                         else if (ino == NILFS_DAT_INO)
1611                                 sc_op = &nilfs_sc_dat_ops;
1612                         else /* file blocks */
1613                                 sc_op = &nilfs_sc_file_ops;
1614                 }
1615                 bh_org = bh;
1616                 get_bh(bh_org);
1617                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1618                                         &binfo);
1619                 if (bh != bh_org)
1620                         nilfs_list_replace_buffer(bh_org, bh);
1621                 brelse(bh_org);
1622                 if (unlikely(err))
1623                         goto failed_bmap;
1624
1625                 if (ndatablk > 0)
1626                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1627                 else
1628                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1629
1630                 blocknr++;
1631                 if (--nblocks == 0) {
1632                         finfo = NULL;
1633                         if (--nfinfo == 0)
1634                                 break;
1635                 } else if (ndatablk > 0)
1636                         ndatablk--;
1637         }
1638  out:
1639         return 0;
1640
1641  failed_bmap:
1642         return err;
1643 }
1644
1645 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1646 {
1647         struct nilfs_segment_buffer *segbuf;
1648         int err;
1649
1650         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1651                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1652                 if (unlikely(err))
1653                         return err;
1654                 nilfs_segbuf_fill_in_segsum(segbuf);
1655         }
1656         return 0;
1657 }
1658
1659 static void nilfs_begin_page_io(struct page *page)
1660 {
1661         if (!page || PageWriteback(page))
1662                 /*
1663                  * For split b-tree node pages, this function may be called
1664                  * twice.  We ignore the 2nd or later calls by this check.
1665                  */
1666                 return;
1667
1668         lock_page(page);
1669         clear_page_dirty_for_io(page);
1670         set_page_writeback(page);
1671         unlock_page(page);
1672 }
1673
1674 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1675 {
1676         struct nilfs_segment_buffer *segbuf;
1677         struct page *bd_page = NULL, *fs_page = NULL;
1678
1679         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1680                 struct buffer_head *bh;
1681
1682                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1683                                     b_assoc_buffers) {
1684                         if (bh->b_page != bd_page) {
1685                                 if (bd_page) {
1686                                         lock_page(bd_page);
1687                                         clear_page_dirty_for_io(bd_page);
1688                                         set_page_writeback(bd_page);
1689                                         unlock_page(bd_page);
1690                                 }
1691                                 bd_page = bh->b_page;
1692                         }
1693                 }
1694
1695                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1696                                     b_assoc_buffers) {
1697                         set_buffer_async_write(bh);
1698                         if (bh == segbuf->sb_super_root) {
1699                                 if (bh->b_page != bd_page) {
1700                                         lock_page(bd_page);
1701                                         clear_page_dirty_for_io(bd_page);
1702                                         set_page_writeback(bd_page);
1703                                         unlock_page(bd_page);
1704                                         bd_page = bh->b_page;
1705                                 }
1706                                 break;
1707                         }
1708                         if (bh->b_page != fs_page) {
1709                                 nilfs_begin_page_io(fs_page);
1710                                 fs_page = bh->b_page;
1711                         }
1712                 }
1713         }
1714         if (bd_page) {
1715                 lock_page(bd_page);
1716                 clear_page_dirty_for_io(bd_page);
1717                 set_page_writeback(bd_page);
1718                 unlock_page(bd_page);
1719         }
1720         nilfs_begin_page_io(fs_page);
1721 }
1722
1723 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1724                                struct the_nilfs *nilfs)
1725 {
1726         int ret;
1727
1728         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1729         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1730         return ret;
1731 }
1732
1733 static void nilfs_end_page_io(struct page *page, int err)
1734 {
1735         if (!page)
1736                 return;
1737
1738         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1739                 /*
1740                  * For b-tree node pages, this function may be called twice
1741                  * or more because they might be split in a segment.
1742                  */
1743                 if (PageDirty(page)) {
1744                         /*
1745                          * For pages holding split b-tree node buffers, dirty
1746                          * flag on the buffers may be cleared discretely.
1747                          * In that case, the page is once redirtied for
1748                          * remaining buffers, and it must be cancelled if
1749                          * all the buffers get cleaned later.
1750                          */
1751                         lock_page(page);
1752                         if (nilfs_page_buffers_clean(page))
1753                                 __nilfs_clear_page_dirty(page);
1754                         unlock_page(page);
1755                 }
1756                 return;
1757         }
1758
1759         if (!err) {
1760                 if (!nilfs_page_buffers_clean(page))
1761                         __set_page_dirty_nobuffers(page);
1762                 ClearPageError(page);
1763         } else {
1764                 __set_page_dirty_nobuffers(page);
1765                 SetPageError(page);
1766         }
1767
1768         end_page_writeback(page);
1769 }
1770
1771 static void nilfs_abort_logs(struct list_head *logs, int err)
1772 {
1773         struct nilfs_segment_buffer *segbuf;
1774         struct page *bd_page = NULL, *fs_page = NULL;
1775         struct buffer_head *bh;
1776
1777         if (list_empty(logs))
1778                 return;
1779
1780         list_for_each_entry(segbuf, logs, sb_list) {
1781                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1782                                     b_assoc_buffers) {
1783                         if (bh->b_page != bd_page) {
1784                                 if (bd_page)
1785                                         end_page_writeback(bd_page);
1786                                 bd_page = bh->b_page;
1787                         }
1788                 }
1789
1790                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1791                                     b_assoc_buffers) {
1792                         clear_buffer_async_write(bh);
1793                         if (bh == segbuf->sb_super_root) {
1794                                 if (bh->b_page != bd_page) {
1795                                         end_page_writeback(bd_page);
1796                                         bd_page = bh->b_page;
1797                                 }
1798                                 break;
1799                         }
1800                         if (bh->b_page != fs_page) {
1801                                 nilfs_end_page_io(fs_page, err);
1802                                 fs_page = bh->b_page;
1803                         }
1804                 }
1805         }
1806         if (bd_page)
1807                 end_page_writeback(bd_page);
1808
1809         nilfs_end_page_io(fs_page, err);
1810 }
1811
1812 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1813                                              struct the_nilfs *nilfs, int err)
1814 {
1815         LIST_HEAD(logs);
1816         int ret;
1817
1818         list_splice_tail_init(&sci->sc_write_logs, &logs);
1819         ret = nilfs_wait_on_logs(&logs);
1820         nilfs_abort_logs(&logs, ret ? : err);
1821
1822         list_splice_tail_init(&sci->sc_segbufs, &logs);
1823         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1824         nilfs_free_incomplete_logs(&logs, nilfs);
1825
1826         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1827                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1828                                                 sci->sc_freesegs,
1829                                                 sci->sc_nfreesegs,
1830                                                 NULL);
1831                 WARN_ON(ret); /* do not happen */
1832         }
1833
1834         nilfs_destroy_logs(&logs);
1835 }
1836
1837 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1838                                    struct nilfs_segment_buffer *segbuf)
1839 {
1840         nilfs->ns_segnum = segbuf->sb_segnum;
1841         nilfs->ns_nextnum = segbuf->sb_nextnum;
1842         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1843                 + segbuf->sb_sum.nblocks;
1844         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1845         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1846 }
1847
1848 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1849 {
1850         struct nilfs_segment_buffer *segbuf;
1851         struct page *bd_page = NULL, *fs_page = NULL;
1852         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1853         int update_sr = false;
1854
1855         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1856                 struct buffer_head *bh;
1857
1858                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1859                                     b_assoc_buffers) {
1860                         set_buffer_uptodate(bh);
1861                         clear_buffer_dirty(bh);
1862                         if (bh->b_page != bd_page) {
1863                                 if (bd_page)
1864                                         end_page_writeback(bd_page);
1865                                 bd_page = bh->b_page;
1866                         }
1867                 }
1868                 /*
1869                  * We assume that the buffers which belong to the same page
1870                  * continue over the buffer list.
1871                  * Under this assumption, the last BHs of pages is
1872                  * identifiable by the discontinuity of bh->b_page
1873                  * (page != fs_page).
1874                  *
1875                  * For B-tree node blocks, however, this assumption is not
1876                  * guaranteed.  The cleanup code of B-tree node pages needs
1877                  * special care.
1878                  */
1879                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1880                                     b_assoc_buffers) {
1881                         const unsigned long set_bits = BIT(BH_Uptodate);
1882                         const unsigned long clear_bits =
1883                                 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1884                                  BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1885                                  BIT(BH_NILFS_Redirected));
1886
1887                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1888                         if (bh == segbuf->sb_super_root) {
1889                                 if (bh->b_page != bd_page) {
1890                                         end_page_writeback(bd_page);
1891                                         bd_page = bh->b_page;
1892                                 }
1893                                 update_sr = true;
1894                                 break;
1895                         }
1896                         if (bh->b_page != fs_page) {
1897                                 nilfs_end_page_io(fs_page, 0);
1898                                 fs_page = bh->b_page;
1899                         }
1900                 }
1901
1902                 if (!nilfs_segbuf_simplex(segbuf)) {
1903                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1904                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1905                                 sci->sc_lseg_stime = jiffies;
1906                         }
1907                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1908                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1909                 }
1910         }
1911         /*
1912          * Since pages may continue over multiple segment buffers,
1913          * end of the last page must be checked outside of the loop.
1914          */
1915         if (bd_page)
1916                 end_page_writeback(bd_page);
1917
1918         nilfs_end_page_io(fs_page, 0);
1919
1920         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1921
1922         if (nilfs_doing_gc())
1923                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1924         else
1925                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1926
1927         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1928
1929         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1930         nilfs_set_next_segment(nilfs, segbuf);
1931
1932         if (update_sr) {
1933                 nilfs->ns_flushed_device = 0;
1934                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1935                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1936
1937                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1938                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1939                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1940                 nilfs_segctor_clear_metadata_dirty(sci);
1941         } else
1942                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1943 }
1944
1945 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1946 {
1947         int ret;
1948
1949         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1950         if (!ret) {
1951                 nilfs_segctor_complete_write(sci);
1952                 nilfs_destroy_logs(&sci->sc_write_logs);
1953         }
1954         return ret;
1955 }
1956
1957 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1958                                              struct the_nilfs *nilfs)
1959 {
1960         struct nilfs_inode_info *ii, *n;
1961         struct inode *ifile = sci->sc_root->ifile;
1962
1963         spin_lock(&nilfs->ns_inode_lock);
1964  retry:
1965         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1966                 if (!ii->i_bh) {
1967                         struct buffer_head *ibh;
1968                         int err;
1969
1970                         spin_unlock(&nilfs->ns_inode_lock);
1971                         err = nilfs_ifile_get_inode_block(
1972                                 ifile, ii->vfs_inode.i_ino, &ibh);
1973                         if (unlikely(err)) {
1974                                 nilfs_warn(sci->sc_super,
1975                                            "log writer: error %d getting inode block (ino=%lu)",
1976                                            err, ii->vfs_inode.i_ino);
1977                                 return err;
1978                         }
1979                         spin_lock(&nilfs->ns_inode_lock);
1980                         if (likely(!ii->i_bh))
1981                                 ii->i_bh = ibh;
1982                         else
1983                                 brelse(ibh);
1984                         goto retry;
1985                 }
1986
1987                 // Always redirty the buffer to avoid race condition
1988                 mark_buffer_dirty(ii->i_bh);
1989                 nilfs_mdt_mark_dirty(ifile);
1990
1991                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1992                 set_bit(NILFS_I_BUSY, &ii->i_state);
1993                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1994         }
1995         spin_unlock(&nilfs->ns_inode_lock);
1996
1997         return 0;
1998 }
1999
2000 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2001                                              struct the_nilfs *nilfs)
2002 {
2003         struct nilfs_inode_info *ii, *n;
2004         int during_mount = !(sci->sc_super->s_flags & SB_ACTIVE);
2005         int defer_iput = false;
2006
2007         spin_lock(&nilfs->ns_inode_lock);
2008         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2009                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2010                     test_bit(NILFS_I_DIRTY, &ii->i_state))
2011                         continue;
2012
2013                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2014                 brelse(ii->i_bh);
2015                 ii->i_bh = NULL;
2016                 list_del_init(&ii->i_dirty);
2017                 if (!ii->vfs_inode.i_nlink || during_mount) {
2018                         /*
2019                          * Defer calling iput() to avoid deadlocks if
2020                          * i_nlink == 0 or mount is not yet finished.
2021                          */
2022                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2023                         defer_iput = true;
2024                 } else {
2025                         spin_unlock(&nilfs->ns_inode_lock);
2026                         iput(&ii->vfs_inode);
2027                         spin_lock(&nilfs->ns_inode_lock);
2028                 }
2029         }
2030         spin_unlock(&nilfs->ns_inode_lock);
2031
2032         if (defer_iput)
2033                 schedule_work(&sci->sc_iput_work);
2034 }
2035
2036 /*
2037  * Main procedure of segment constructor
2038  */
2039 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2040 {
2041         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2042         int err;
2043
2044         if (sb_rdonly(sci->sc_super))
2045                 return -EROFS;
2046
2047         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2048         sci->sc_cno = nilfs->ns_cno;
2049
2050         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2051         if (unlikely(err))
2052                 goto out;
2053
2054         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2055                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2056
2057         if (nilfs_segctor_clean(sci))
2058                 goto out;
2059
2060         do {
2061                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2062
2063                 err = nilfs_segctor_begin_construction(sci, nilfs);
2064                 if (unlikely(err))
2065                         goto out;
2066
2067                 /* Update time stamp */
2068                 sci->sc_seg_ctime = ktime_get_real_seconds();
2069
2070                 err = nilfs_segctor_collect(sci, nilfs, mode);
2071                 if (unlikely(err))
2072                         goto failed;
2073
2074                 /* Avoid empty segment */
2075                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2076                     nilfs_segbuf_empty(sci->sc_curseg)) {
2077                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2078                         goto out;
2079                 }
2080
2081                 err = nilfs_segctor_assign(sci, mode);
2082                 if (unlikely(err))
2083                         goto failed;
2084
2085                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2086                         nilfs_segctor_fill_in_file_bmap(sci);
2087
2088                 if (mode == SC_LSEG_SR &&
2089                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2090                         err = nilfs_segctor_fill_in_checkpoint(sci);
2091                         if (unlikely(err))
2092                                 goto failed_to_write;
2093
2094                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2095                 }
2096                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2097
2098                 /* Write partial segments */
2099                 nilfs_segctor_prepare_write(sci);
2100
2101                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2102                                             nilfs->ns_crc_seed);
2103
2104                 err = nilfs_segctor_write(sci, nilfs);
2105                 if (unlikely(err))
2106                         goto failed_to_write;
2107
2108                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2109                     nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2110                         /*
2111                          * At this point, we avoid double buffering
2112                          * for blocksize < pagesize because page dirty
2113                          * flag is turned off during write and dirty
2114                          * buffers are not properly collected for
2115                          * pages crossing over segments.
2116                          */
2117                         err = nilfs_segctor_wait(sci);
2118                         if (err)
2119                                 goto failed_to_write;
2120                 }
2121         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2122
2123  out:
2124         nilfs_segctor_drop_written_files(sci, nilfs);
2125         return err;
2126
2127  failed_to_write:
2128         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2129                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2130
2131  failed:
2132         if (nilfs_doing_gc())
2133                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2134         nilfs_segctor_abort_construction(sci, nilfs, err);
2135         goto out;
2136 }
2137
2138 /**
2139  * nilfs_segctor_start_timer - set timer of background write
2140  * @sci: nilfs_sc_info
2141  *
2142  * If the timer has already been set, it ignores the new request.
2143  * This function MUST be called within a section locking the segment
2144  * semaphore.
2145  */
2146 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2147 {
2148         spin_lock(&sci->sc_state_lock);
2149         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2150                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2151                 add_timer(&sci->sc_timer);
2152                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2153         }
2154         spin_unlock(&sci->sc_state_lock);
2155 }
2156
2157 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2158 {
2159         spin_lock(&sci->sc_state_lock);
2160         if (!(sci->sc_flush_request & BIT(bn))) {
2161                 unsigned long prev_req = sci->sc_flush_request;
2162
2163                 sci->sc_flush_request |= BIT(bn);
2164                 if (!prev_req)
2165                         wake_up(&sci->sc_wait_daemon);
2166         }
2167         spin_unlock(&sci->sc_state_lock);
2168 }
2169
2170 /**
2171  * nilfs_flush_segment - trigger a segment construction for resource control
2172  * @sb: super block
2173  * @ino: inode number of the file to be flushed out.
2174  */
2175 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2176 {
2177         struct the_nilfs *nilfs = sb->s_fs_info;
2178         struct nilfs_sc_info *sci = nilfs->ns_writer;
2179
2180         if (!sci || nilfs_doing_construction())
2181                 return;
2182         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2183                                         /* assign bit 0 to data files */
2184 }
2185
2186 struct nilfs_segctor_wait_request {
2187         wait_queue_entry_t      wq;
2188         __u32           seq;
2189         int             err;
2190         atomic_t        done;
2191 };
2192
2193 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2194 {
2195         struct nilfs_segctor_wait_request wait_req;
2196         int err = 0;
2197
2198         spin_lock(&sci->sc_state_lock);
2199         init_wait(&wait_req.wq);
2200         wait_req.err = 0;
2201         atomic_set(&wait_req.done, 0);
2202         wait_req.seq = ++sci->sc_seq_request;
2203         spin_unlock(&sci->sc_state_lock);
2204
2205         init_waitqueue_entry(&wait_req.wq, current);
2206         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2207         set_current_state(TASK_INTERRUPTIBLE);
2208         wake_up(&sci->sc_wait_daemon);
2209
2210         for (;;) {
2211                 if (atomic_read(&wait_req.done)) {
2212                         err = wait_req.err;
2213                         break;
2214                 }
2215                 if (!signal_pending(current)) {
2216                         schedule();
2217                         continue;
2218                 }
2219                 err = -ERESTARTSYS;
2220                 break;
2221         }
2222         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2223         return err;
2224 }
2225
2226 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2227 {
2228         struct nilfs_segctor_wait_request *wrq, *n;
2229         unsigned long flags;
2230
2231         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2232         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2233                 if (!atomic_read(&wrq->done) &&
2234                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2235                         wrq->err = err;
2236                         atomic_set(&wrq->done, 1);
2237                 }
2238                 if (atomic_read(&wrq->done)) {
2239                         wrq->wq.func(&wrq->wq,
2240                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2241                                      0, NULL);
2242                 }
2243         }
2244         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2245 }
2246
2247 /**
2248  * nilfs_construct_segment - construct a logical segment
2249  * @sb: super block
2250  *
2251  * Return Value: On success, 0 is returned. On errors, one of the following
2252  * negative error code is returned.
2253  *
2254  * %-EROFS - Read only filesystem.
2255  *
2256  * %-EIO - I/O error
2257  *
2258  * %-ENOSPC - No space left on device (only in a panic state).
2259  *
2260  * %-ERESTARTSYS - Interrupted.
2261  *
2262  * %-ENOMEM - Insufficient memory available.
2263  */
2264 int nilfs_construct_segment(struct super_block *sb)
2265 {
2266         struct the_nilfs *nilfs = sb->s_fs_info;
2267         struct nilfs_sc_info *sci = nilfs->ns_writer;
2268         struct nilfs_transaction_info *ti;
2269
2270         if (sb_rdonly(sb) || unlikely(!sci))
2271                 return -EROFS;
2272
2273         /* A call inside transactions causes a deadlock. */
2274         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2275
2276         return nilfs_segctor_sync(sci);
2277 }
2278
2279 /**
2280  * nilfs_construct_dsync_segment - construct a data-only logical segment
2281  * @sb: super block
2282  * @inode: inode whose data blocks should be written out
2283  * @start: start byte offset
2284  * @end: end byte offset (inclusive)
2285  *
2286  * Return Value: On success, 0 is returned. On errors, one of the following
2287  * negative error code is returned.
2288  *
2289  * %-EROFS - Read only filesystem.
2290  *
2291  * %-EIO - I/O error
2292  *
2293  * %-ENOSPC - No space left on device (only in a panic state).
2294  *
2295  * %-ERESTARTSYS - Interrupted.
2296  *
2297  * %-ENOMEM - Insufficient memory available.
2298  */
2299 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2300                                   loff_t start, loff_t end)
2301 {
2302         struct the_nilfs *nilfs = sb->s_fs_info;
2303         struct nilfs_sc_info *sci = nilfs->ns_writer;
2304         struct nilfs_inode_info *ii;
2305         struct nilfs_transaction_info ti;
2306         int err = 0;
2307
2308         if (sb_rdonly(sb) || unlikely(!sci))
2309                 return -EROFS;
2310
2311         nilfs_transaction_lock(sb, &ti, 0);
2312
2313         ii = NILFS_I(inode);
2314         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2315             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2316             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2317             nilfs_discontinued(nilfs)) {
2318                 nilfs_transaction_unlock(sb);
2319                 err = nilfs_segctor_sync(sci);
2320                 return err;
2321         }
2322
2323         spin_lock(&nilfs->ns_inode_lock);
2324         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2325             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2326                 spin_unlock(&nilfs->ns_inode_lock);
2327                 nilfs_transaction_unlock(sb);
2328                 return 0;
2329         }
2330         spin_unlock(&nilfs->ns_inode_lock);
2331         sci->sc_dsync_inode = ii;
2332         sci->sc_dsync_start = start;
2333         sci->sc_dsync_end = end;
2334
2335         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2336         if (!err)
2337                 nilfs->ns_flushed_device = 0;
2338
2339         nilfs_transaction_unlock(sb);
2340         return err;
2341 }
2342
2343 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2344 #define FLUSH_DAT_BIT   BIT(NILFS_DAT_INO) /* DAT only */
2345
2346 /**
2347  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2348  * @sci: segment constructor object
2349  */
2350 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2351 {
2352         spin_lock(&sci->sc_state_lock);
2353         sci->sc_seq_accepted = sci->sc_seq_request;
2354         spin_unlock(&sci->sc_state_lock);
2355         del_timer_sync(&sci->sc_timer);
2356 }
2357
2358 /**
2359  * nilfs_segctor_notify - notify the result of request to caller threads
2360  * @sci: segment constructor object
2361  * @mode: mode of log forming
2362  * @err: error code to be notified
2363  */
2364 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2365 {
2366         /* Clear requests (even when the construction failed) */
2367         spin_lock(&sci->sc_state_lock);
2368
2369         if (mode == SC_LSEG_SR) {
2370                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2371                 sci->sc_seq_done = sci->sc_seq_accepted;
2372                 nilfs_segctor_wakeup(sci, err);
2373                 sci->sc_flush_request = 0;
2374         } else {
2375                 if (mode == SC_FLUSH_FILE)
2376                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2377                 else if (mode == SC_FLUSH_DAT)
2378                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2379
2380                 /* re-enable timer if checkpoint creation was not done */
2381                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2382                     time_before(jiffies, sci->sc_timer.expires))
2383                         add_timer(&sci->sc_timer);
2384         }
2385         spin_unlock(&sci->sc_state_lock);
2386 }
2387
2388 /**
2389  * nilfs_segctor_construct - form logs and write them to disk
2390  * @sci: segment constructor object
2391  * @mode: mode of log forming
2392  */
2393 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2394 {
2395         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2396         struct nilfs_super_block **sbp;
2397         int err = 0;
2398
2399         nilfs_segctor_accept(sci);
2400
2401         if (nilfs_discontinued(nilfs))
2402                 mode = SC_LSEG_SR;
2403         if (!nilfs_segctor_confirm(sci))
2404                 err = nilfs_segctor_do_construct(sci, mode);
2405
2406         if (likely(!err)) {
2407                 if (mode != SC_FLUSH_DAT)
2408                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2409                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2410                     nilfs_discontinued(nilfs)) {
2411                         down_write(&nilfs->ns_sem);
2412                         err = -EIO;
2413                         sbp = nilfs_prepare_super(sci->sc_super,
2414                                                   nilfs_sb_will_flip(nilfs));
2415                         if (likely(sbp)) {
2416                                 nilfs_set_log_cursor(sbp[0], nilfs);
2417                                 err = nilfs_commit_super(sci->sc_super,
2418                                                          NILFS_SB_COMMIT);
2419                         }
2420                         up_write(&nilfs->ns_sem);
2421                 }
2422         }
2423
2424         nilfs_segctor_notify(sci, mode, err);
2425         return err;
2426 }
2427
2428 static void nilfs_construction_timeout(struct timer_list *t)
2429 {
2430         struct nilfs_sc_info *sci = from_timer(sci, t, sc_timer);
2431
2432         wake_up_process(sci->sc_timer_task);
2433 }
2434
2435 static void
2436 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2437 {
2438         struct nilfs_inode_info *ii, *n;
2439
2440         list_for_each_entry_safe(ii, n, head, i_dirty) {
2441                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2442                         continue;
2443                 list_del_init(&ii->i_dirty);
2444                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2445                 nilfs_btnode_cache_clear(ii->i_assoc_inode->i_mapping);
2446                 iput(&ii->vfs_inode);
2447         }
2448 }
2449
2450 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2451                          void **kbufs)
2452 {
2453         struct the_nilfs *nilfs = sb->s_fs_info;
2454         struct nilfs_sc_info *sci = nilfs->ns_writer;
2455         struct nilfs_transaction_info ti;
2456         int err;
2457
2458         if (unlikely(!sci))
2459                 return -EROFS;
2460
2461         nilfs_transaction_lock(sb, &ti, 1);
2462
2463         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2464         if (unlikely(err))
2465                 goto out_unlock;
2466
2467         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2468         if (unlikely(err)) {
2469                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2470                 goto out_unlock;
2471         }
2472
2473         sci->sc_freesegs = kbufs[4];
2474         sci->sc_nfreesegs = argv[4].v_nmembs;
2475         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2476
2477         for (;;) {
2478                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2479                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2480
2481                 if (likely(!err))
2482                         break;
2483
2484                 nilfs_warn(sb, "error %d cleaning segments", err);
2485                 set_current_state(TASK_INTERRUPTIBLE);
2486                 schedule_timeout(sci->sc_interval);
2487         }
2488         if (nilfs_test_opt(nilfs, DISCARD)) {
2489                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2490                                                  sci->sc_nfreesegs);
2491                 if (ret) {
2492                         nilfs_warn(sb,
2493                                    "error %d on discard request, turning discards off for the device",
2494                                    ret);
2495                         nilfs_clear_opt(nilfs, DISCARD);
2496                 }
2497         }
2498
2499  out_unlock:
2500         sci->sc_freesegs = NULL;
2501         sci->sc_nfreesegs = 0;
2502         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2503         nilfs_transaction_unlock(sb);
2504         return err;
2505 }
2506
2507 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2508 {
2509         struct nilfs_transaction_info ti;
2510
2511         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2512         nilfs_segctor_construct(sci, mode);
2513
2514         /*
2515          * Unclosed segment should be retried.  We do this using sc_timer.
2516          * Timeout of sc_timer will invoke complete construction which leads
2517          * to close the current logical segment.
2518          */
2519         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2520                 nilfs_segctor_start_timer(sci);
2521
2522         nilfs_transaction_unlock(sci->sc_super);
2523 }
2524
2525 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2526 {
2527         int mode = 0;
2528
2529         spin_lock(&sci->sc_state_lock);
2530         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2531                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2532         spin_unlock(&sci->sc_state_lock);
2533
2534         if (mode) {
2535                 nilfs_segctor_do_construct(sci, mode);
2536
2537                 spin_lock(&sci->sc_state_lock);
2538                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2539                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2540                 spin_unlock(&sci->sc_state_lock);
2541         }
2542         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2543 }
2544
2545 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2546 {
2547         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2548             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2549                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2550                         return SC_FLUSH_FILE;
2551                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2552                         return SC_FLUSH_DAT;
2553         }
2554         return SC_LSEG_SR;
2555 }
2556
2557 /**
2558  * nilfs_segctor_thread - main loop of the segment constructor thread.
2559  * @arg: pointer to a struct nilfs_sc_info.
2560  *
2561  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2562  * to execute segment constructions.
2563  */
2564 static int nilfs_segctor_thread(void *arg)
2565 {
2566         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2567         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2568         int timeout = 0;
2569
2570         sci->sc_timer_task = current;
2571
2572         /* start sync. */
2573         sci->sc_task = current;
2574         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2575         nilfs_info(sci->sc_super,
2576                    "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2577                    sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2578
2579         spin_lock(&sci->sc_state_lock);
2580  loop:
2581         for (;;) {
2582                 int mode;
2583
2584                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2585                         goto end_thread;
2586
2587                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2588                         mode = SC_LSEG_SR;
2589                 else if (sci->sc_flush_request)
2590                         mode = nilfs_segctor_flush_mode(sci);
2591                 else
2592                         break;
2593
2594                 spin_unlock(&sci->sc_state_lock);
2595                 nilfs_segctor_thread_construct(sci, mode);
2596                 spin_lock(&sci->sc_state_lock);
2597                 timeout = 0;
2598         }
2599
2600
2601         if (freezing(current)) {
2602                 spin_unlock(&sci->sc_state_lock);
2603                 try_to_freeze();
2604                 spin_lock(&sci->sc_state_lock);
2605         } else {
2606                 DEFINE_WAIT(wait);
2607                 int should_sleep = 1;
2608
2609                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2610                                 TASK_INTERRUPTIBLE);
2611
2612                 if (sci->sc_seq_request != sci->sc_seq_done)
2613                         should_sleep = 0;
2614                 else if (sci->sc_flush_request)
2615                         should_sleep = 0;
2616                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2617                         should_sleep = time_before(jiffies,
2618                                         sci->sc_timer.expires);
2619
2620                 if (should_sleep) {
2621                         spin_unlock(&sci->sc_state_lock);
2622                         schedule();
2623                         spin_lock(&sci->sc_state_lock);
2624                 }
2625                 finish_wait(&sci->sc_wait_daemon, &wait);
2626                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2627                            time_after_eq(jiffies, sci->sc_timer.expires));
2628
2629                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2630                         set_nilfs_discontinued(nilfs);
2631         }
2632         goto loop;
2633
2634  end_thread:
2635         /* end sync. */
2636         sci->sc_task = NULL;
2637         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2638         spin_unlock(&sci->sc_state_lock);
2639         return 0;
2640 }
2641
2642 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2643 {
2644         struct task_struct *t;
2645
2646         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2647         if (IS_ERR(t)) {
2648                 int err = PTR_ERR(t);
2649
2650                 nilfs_err(sci->sc_super, "error %d creating segctord thread",
2651                           err);
2652                 return err;
2653         }
2654         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2655         return 0;
2656 }
2657
2658 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2659         __acquires(&sci->sc_state_lock)
2660         __releases(&sci->sc_state_lock)
2661 {
2662         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2663
2664         while (sci->sc_task) {
2665                 wake_up(&sci->sc_wait_daemon);
2666                 spin_unlock(&sci->sc_state_lock);
2667                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2668                 spin_lock(&sci->sc_state_lock);
2669         }
2670 }
2671
2672 /*
2673  * Setup & clean-up functions
2674  */
2675 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2676                                                struct nilfs_root *root)
2677 {
2678         struct the_nilfs *nilfs = sb->s_fs_info;
2679         struct nilfs_sc_info *sci;
2680
2681         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2682         if (!sci)
2683                 return NULL;
2684
2685         sci->sc_super = sb;
2686
2687         nilfs_get_root(root);
2688         sci->sc_root = root;
2689
2690         init_waitqueue_head(&sci->sc_wait_request);
2691         init_waitqueue_head(&sci->sc_wait_daemon);
2692         init_waitqueue_head(&sci->sc_wait_task);
2693         spin_lock_init(&sci->sc_state_lock);
2694         INIT_LIST_HEAD(&sci->sc_dirty_files);
2695         INIT_LIST_HEAD(&sci->sc_segbufs);
2696         INIT_LIST_HEAD(&sci->sc_write_logs);
2697         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2698         INIT_LIST_HEAD(&sci->sc_iput_queue);
2699         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2700         timer_setup(&sci->sc_timer, nilfs_construction_timeout, 0);
2701
2702         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2703         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2704         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2705
2706         if (nilfs->ns_interval)
2707                 sci->sc_interval = HZ * nilfs->ns_interval;
2708         if (nilfs->ns_watermark)
2709                 sci->sc_watermark = nilfs->ns_watermark;
2710         return sci;
2711 }
2712
2713 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2714 {
2715         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2716
2717         /*
2718          * The segctord thread was stopped and its timer was removed.
2719          * But some tasks remain.
2720          */
2721         do {
2722                 struct nilfs_transaction_info ti;
2723
2724                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2725                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2726                 nilfs_transaction_unlock(sci->sc_super);
2727
2728                 flush_work(&sci->sc_iput_work);
2729
2730         } while (ret && ret != -EROFS && retrycount-- > 0);
2731 }
2732
2733 /**
2734  * nilfs_segctor_destroy - destroy the segment constructor.
2735  * @sci: nilfs_sc_info
2736  *
2737  * nilfs_segctor_destroy() kills the segctord thread and frees
2738  * the nilfs_sc_info struct.
2739  * Caller must hold the segment semaphore.
2740  */
2741 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2742 {
2743         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2744         int flag;
2745
2746         up_write(&nilfs->ns_segctor_sem);
2747
2748         spin_lock(&sci->sc_state_lock);
2749         nilfs_segctor_kill_thread(sci);
2750         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2751                 || sci->sc_seq_request != sci->sc_seq_done);
2752         spin_unlock(&sci->sc_state_lock);
2753
2754         if (flush_work(&sci->sc_iput_work))
2755                 flag = true;
2756
2757         if (flag || !nilfs_segctor_confirm(sci))
2758                 nilfs_segctor_write_out(sci);
2759
2760         if (!list_empty(&sci->sc_dirty_files)) {
2761                 nilfs_warn(sci->sc_super,
2762                            "disposed unprocessed dirty file(s) when stopping log writer");
2763                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2764         }
2765
2766         if (!list_empty(&sci->sc_iput_queue)) {
2767                 nilfs_warn(sci->sc_super,
2768                            "disposed unprocessed inode(s) in iput queue when stopping log writer");
2769                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2770         }
2771
2772         WARN_ON(!list_empty(&sci->sc_segbufs));
2773         WARN_ON(!list_empty(&sci->sc_write_logs));
2774
2775         nilfs_put_root(sci->sc_root);
2776
2777         down_write(&nilfs->ns_segctor_sem);
2778
2779         timer_shutdown_sync(&sci->sc_timer);
2780         kfree(sci);
2781 }
2782
2783 /**
2784  * nilfs_attach_log_writer - attach log writer
2785  * @sb: super block instance
2786  * @root: root object of the current filesystem tree
2787  *
2788  * This allocates a log writer object, initializes it, and starts the
2789  * log writer.
2790  *
2791  * Return Value: On success, 0 is returned. On error, one of the following
2792  * negative error code is returned.
2793  *
2794  * %-ENOMEM - Insufficient memory available.
2795  */
2796 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2797 {
2798         struct the_nilfs *nilfs = sb->s_fs_info;
2799         int err;
2800
2801         if (nilfs->ns_writer) {
2802                 /*
2803                  * This happens if the filesystem is made read-only by
2804                  * __nilfs_error or nilfs_remount and then remounted
2805                  * read/write.  In these cases, reuse the existing
2806                  * writer.
2807                  */
2808                 return 0;
2809         }
2810
2811         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2812         if (!nilfs->ns_writer)
2813                 return -ENOMEM;
2814
2815         inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2816
2817         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2818         if (unlikely(err))
2819                 nilfs_detach_log_writer(sb);
2820
2821         return err;
2822 }
2823
2824 /**
2825  * nilfs_detach_log_writer - destroy log writer
2826  * @sb: super block instance
2827  *
2828  * This kills log writer daemon, frees the log writer object, and
2829  * destroys list of dirty files.
2830  */
2831 void nilfs_detach_log_writer(struct super_block *sb)
2832 {
2833         struct the_nilfs *nilfs = sb->s_fs_info;
2834         LIST_HEAD(garbage_list);
2835
2836         down_write(&nilfs->ns_segctor_sem);
2837         if (nilfs->ns_writer) {
2838                 nilfs_segctor_destroy(nilfs->ns_writer);
2839                 nilfs->ns_writer = NULL;
2840         }
2841
2842         /* Force to free the list of dirty files */
2843         spin_lock(&nilfs->ns_inode_lock);
2844         if (!list_empty(&nilfs->ns_dirty_files)) {
2845                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2846                 nilfs_warn(sb,
2847                            "disposed unprocessed dirty file(s) when detaching log writer");
2848         }
2849         spin_unlock(&nilfs->ns_inode_lock);
2850         up_write(&nilfs->ns_segctor_sem);
2851
2852         nilfs_dispose_list(nilfs, &garbage_list, 1);
2853 }