fs/nilfs2/segment.c

   1 /*
   2  * segment.c - NILFS segment constructor.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/pagemap.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/writeback.h>
  27 #include <linux/bio.h>
  28 #include <linux/completion.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/freezer.h>
  32 #include <linux/kthread.h>
  33 #include <linux/crc32.h>
  34 #include <linux/pagevec.h>
  35 #include <linux/slab.h>
  36 #include "nilfs.h"
  37 #include "btnode.h"
  38 #include "page.h"
  39 #include "segment.h"
  40 #include "sufile.h"
  41 #include "cpfile.h"
  42 #include "ifile.h"
  43 #include "segbuf.h"
  44
  45
  46 /*
  47  * Segment constructor
  48  */
  49 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  50
  51 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
  52                                 appended in collection retry loop */
  53
  54 /* Construction mode */
  55 enum {
  56         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  57         SC_LSEG_DSYNC,  /* Flush data blocks of a given file and make
  58                            a logical segment without a super root */
  59         SC_FLUSH_FILE,  /* Flush data files, leads to segment writes without
  60                            creating a checkpoint */
  61         SC_FLUSH_DAT,   /* Flush DAT file. This also creates segments without
  62                            a checkpoint */
  63 };
  64
  65 /* Stage numbers of dirty block collection */
  66 enum {
  67         NILFS_ST_INIT = 0,
  68         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  69         NILFS_ST_FILE,
  70         NILFS_ST_IFILE,
  71         NILFS_ST_CPFILE,
  72         NILFS_ST_SUFILE,
  73         NILFS_ST_DAT,
  74         NILFS_ST_SR,            /* Super root */
  75         NILFS_ST_DSYNC,         /* Data sync blocks */
  76         NILFS_ST_DONE,
  77 };
  78
  79 /* State flags of collection */
  80 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
  81 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
  82 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
  83 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
  84
  85 /* Operations depending on the construction mode and file type */
  86 struct nilfs_sc_operations {
  87         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
  88                             struct inode *);
  89         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
  90                             struct inode *);
  91         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
  92                             struct inode *);
  93         void (*write_data_binfo)(struct nilfs_sc_info *,
  94                                  struct nilfs_segsum_pointer *,
  95                                  union nilfs_binfo *);
  96         void (*write_node_binfo)(struct nilfs_sc_info *,
  97                                  struct nilfs_segsum_pointer *,
  98                                  union nilfs_binfo *);
  99 };
 100
 101 /*
 102  * Other definitions
 103  */
 104 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 105 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 107 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
 108
 109 #define nilfs_cnt32_gt(a, b)   \
 110         (typecheck(__u32, a) && typecheck(__u32, b) && \
 111          ((__s32)(b) - (__s32)(a) < 0))
 112 #define nilfs_cnt32_ge(a, b)   \
 113         (typecheck(__u32, a) && typecheck(__u32, b) && \
 114          ((__s32)(a) - (__s32)(b) >= 0))
 115 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 116 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 117
 118 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
 119 {
 120         struct nilfs_transaction_info *cur_ti = current->journal_info;
 121         void *save = NULL;
 122
 123         if (cur_ti) {
 124                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 125                         return ++cur_ti->ti_count;
 126                 else {
 127                         /*
 128                          * If journal_info field is occupied by other FS,
 129                          * it is saved and will be restored on
 130                          * nilfs_transaction_commit().
 131                          */
 132                         printk(KERN_WARNING
 133                                "NILFS warning: journal info from a different "
 134                                "FS\n");
 135                         save = current->journal_info;
 136                 }
 137         }
 138         if (!ti) {
 139                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 140                 if (!ti)
 141                         return -ENOMEM;
 142                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 143         } else {
 144                 ti->ti_flags = 0;
 145         }
 146         ti->ti_count = 0;
 147         ti->ti_save = save;
 148         ti->ti_magic = NILFS_TI_MAGIC;
 149         current->journal_info = ti;
 150         return 0;
 151 }
 152
 153 /**
 154  * nilfs_transaction_begin - start indivisible file operations.
 155  * @sb: super block
 156  * @ti: nilfs_transaction_info
 157  * @vacancy_check: flags for vacancy rate checks
 158  *
 159  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 160  * the segment semaphore, to make a segment construction and write tasks
 161  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 162  * The region enclosed by these two functions can be nested.  To avoid a
 163  * deadlock, the semaphore is only acquired or released in the outermost call.
 164  *
 165  * This function allocates a nilfs_transaction_info struct to keep context
 166  * information on it.  It is initialized and hooked onto the current task in
 167  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 168  * instead; otherwise a new struct is assigned from a slab.
 169  *
 170  * When @vacancy_check flag is set, this function will check the amount of
 171  * free space, and will wait for the GC to reclaim disk space if low capacity.
 172  *
 173  * Return Value: On success, 0 is returned. On error, one of the following
 174  * negative error code is returned.
 175  *
 176  * %-ENOMEM - Insufficient memory available.
 177  *
 178  * %-ENOSPC - No space left on device
 179  */
 180 int nilfs_transaction_begin(struct super_block *sb,
 181                             struct nilfs_transaction_info *ti,
 182                             int vacancy_check)
 183 {
 184         struct the_nilfs *nilfs;
 185         int ret = nilfs_prepare_segment_lock(ti);
 186
 187         if (unlikely(ret < 0))
 188                 return ret;
 189         if (ret > 0)
 190                 return 0;
 191
 192         vfs_check_frozen(sb, SB_FREEZE_WRITE);
 193
 194         nilfs = sb->s_fs_info;
 195         down_read(&nilfs->ns_segctor_sem);
 196         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 197                 up_read(&nilfs->ns_segctor_sem);
 198                 ret = -ENOSPC;
 199                 goto failed;
 200         }
 201         return 0;
 202
 203  failed:
 204         ti = current->journal_info;
 205         current->journal_info = ti->ti_save;
 206         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 207                 kmem_cache_free(nilfs_transaction_cachep, ti);
 208         return ret;
 209 }
 210
 211 /**
 212  * nilfs_transaction_commit - commit indivisible file operations.
 213  * @sb: super block
 214  *
 215  * nilfs_transaction_commit() releases the read semaphore which is
 216  * acquired by nilfs_transaction_begin(). This is only performed
 217  * in outermost call of this function.  If a commit flag is set,
 218  * nilfs_transaction_commit() sets a timer to start the segment
 219  * constructor.  If a sync flag is set, it starts construction
 220  * directly.
 221  */
 222 int nilfs_transaction_commit(struct super_block *sb)
 223 {
 224         struct nilfs_transaction_info *ti = current->journal_info;
 225         struct the_nilfs *nilfs = sb->s_fs_info;
 226         int err = 0;
 227
 228         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 229         ti->ti_flags |= NILFS_TI_COMMIT;
 230         if (ti->ti_count > 0) {
 231                 ti->ti_count--;
 232                 return 0;
 233         }
 234         if (nilfs->ns_writer) {
 235                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 236
 237                 if (ti->ti_flags & NILFS_TI_COMMIT)
 238                         nilfs_segctor_start_timer(sci);
 239                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 240                         nilfs_segctor_do_flush(sci, 0);
 241         }
 242         up_read(&nilfs->ns_segctor_sem);
 243         current->journal_info = ti->ti_save;
 244
 245         if (ti->ti_flags & NILFS_TI_SYNC)
 246                 err = nilfs_construct_segment(sb);
 247         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 248                 kmem_cache_free(nilfs_transaction_cachep, ti);
 249         return err;
 250 }
 251
 252 void nilfs_transaction_abort(struct super_block *sb)
 253 {
 254         struct nilfs_transaction_info *ti = current->journal_info;
 255         struct the_nilfs *nilfs = sb->s_fs_info;
 256
 257         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 258         if (ti->ti_count > 0) {
 259                 ti->ti_count--;
 260                 return;
 261         }
 262         up_read(&nilfs->ns_segctor_sem);
 263
 264         current->journal_info = ti->ti_save;
 265         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 266                 kmem_cache_free(nilfs_transaction_cachep, ti);
 267 }
 268
 269 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 270 {
 271         struct the_nilfs *nilfs = sb->s_fs_info;
 272         struct nilfs_sc_info *sci = nilfs->ns_writer;
 273
 274         if (!sci || !sci->sc_flush_request)
 275                 return;
 276
 277         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 278         up_read(&nilfs->ns_segctor_sem);
 279
 280         down_write(&nilfs->ns_segctor_sem);
 281         if (sci->sc_flush_request &&
 282             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 283                 struct nilfs_transaction_info *ti = current->journal_info;
 284
 285                 ti->ti_flags |= NILFS_TI_WRITER;
 286                 nilfs_segctor_do_immediate_flush(sci);
 287                 ti->ti_flags &= ~NILFS_TI_WRITER;
 288         }
 289         downgrade_write(&nilfs->ns_segctor_sem);
 290 }
 291
 292 static void nilfs_transaction_lock(struct super_block *sb,
 293                                    struct nilfs_transaction_info *ti,
 294                                    int gcflag)
 295 {
 296         struct nilfs_transaction_info *cur_ti = current->journal_info;
 297         struct the_nilfs *nilfs = sb->s_fs_info;
 298         struct nilfs_sc_info *sci = nilfs->ns_writer;
 299
 300         WARN_ON(cur_ti);
 301         ti->ti_flags = NILFS_TI_WRITER;
 302         ti->ti_count = 0;
 303         ti->ti_save = cur_ti;
 304         ti->ti_magic = NILFS_TI_MAGIC;
 305         current->journal_info = ti;
 306
 307         for (;;) {
 308                 down_write(&nilfs->ns_segctor_sem);
 309                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 310                         break;
 311
 312                 nilfs_segctor_do_immediate_flush(sci);
 313
 314                 up_write(&nilfs->ns_segctor_sem);
 315                 yield();
 316         }
 317         if (gcflag)
 318                 ti->ti_flags |= NILFS_TI_GC;
 319 }
 320
 321 static void nilfs_transaction_unlock(struct super_block *sb)
 322 {
 323         struct nilfs_transaction_info *ti = current->journal_info;
 324         struct the_nilfs *nilfs = sb->s_fs_info;
 325
 326         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 327         BUG_ON(ti->ti_count > 0);
 328
 329         up_write(&nilfs->ns_segctor_sem);
 330         current->journal_info = ti->ti_save;
 331 }
 332
 333 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 334                                             struct nilfs_segsum_pointer *ssp,
 335                                             unsigned bytes)
 336 {
 337         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 338         unsigned blocksize = sci->sc_super->s_blocksize;
 339         void *p;
 340
 341         if (unlikely(ssp->offset + bytes > blocksize)) {
 342                 ssp->offset = 0;
 343                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 344                                                &segbuf->sb_segsum_buffers));
 345                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 346         }
 347         p = ssp->bh->b_data + ssp->offset;
 348         ssp->offset += bytes;
 349         return p;
 350 }
 351
 352 /**
 353  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 354  * @sci: nilfs_sc_info
 355  */
 356 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 357 {
 358         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 359         struct buffer_head *sumbh;
 360         unsigned sumbytes;
 361         unsigned flags = 0;
 362         int err;
 363
 364         if (nilfs_doing_gc())
 365                 flags = NILFS_SS_GC;
 366         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 367         if (unlikely(err))
 368                 return err;
 369
 370         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 371         sumbytes = segbuf->sb_sum.sumbytes;
 372         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 373         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 374         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 375         return 0;
 376 }
 377
 378 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 379 {
 380         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 381         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 382                 return -E2BIG; /* The current segment is filled up
 383                                   (internal code) */
 384         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 385         return nilfs_segctor_reset_segment_buffer(sci);
 386 }
 387
 388 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 389 {
 390         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 391         int err;
 392
 393         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 394                 err = nilfs_segctor_feed_segment(sci);
 395                 if (err)
 396                         return err;
 397                 segbuf = sci->sc_curseg;
 398         }
 399         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 400         if (likely(!err))
 401                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 402         return err;
 403 }
 404
 405 /*
 406  * Functions for making segment summary and payloads
 407  */
 408 static int nilfs_segctor_segsum_block_required(
 409         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 410         unsigned binfo_size)
 411 {
 412         unsigned blocksize = sci->sc_super->s_blocksize;
 413         /* Size of finfo and binfo is enough small against blocksize */
 414
 415         return ssp->offset + binfo_size +
 416                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 417                 blocksize;
 418 }
 419
 420 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 421                                       struct inode *inode)
 422 {
 423         sci->sc_curseg->sb_sum.nfinfo++;
 424         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 425         nilfs_segctor_map_segsum_entry(
 426                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 427
 428         if (NILFS_I(inode)->i_root &&
 429             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 430                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 431         /* skip finfo */
 432 }
 433
 434 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 435                                     struct inode *inode)
 436 {
 437         struct nilfs_finfo *finfo;
 438         struct nilfs_inode_info *ii;
 439         struct nilfs_segment_buffer *segbuf;
 440         __u64 cno;
 441
 442         if (sci->sc_blk_cnt == 0)
 443                 return;
 444
 445         ii = NILFS_I(inode);
 446
 447         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 448                 cno = ii->i_cno;
 449         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 450                 cno = 0;
 451         else
 452                 cno = sci->sc_cno;
 453
 454         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 455                                                  sizeof(*finfo));
 456         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 457         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 458         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 459         finfo->fi_cno = cpu_to_le64(cno);
 460
 461         segbuf = sci->sc_curseg;
 462         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 463                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 464         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 465         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 466 }
 467
 468 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 469                                         struct buffer_head *bh,
 470                                         struct inode *inode,
 471                                         unsigned binfo_size)
 472 {
 473         struct nilfs_segment_buffer *segbuf;
 474         int required, err = 0;
 475
 476  retry:
 477         segbuf = sci->sc_curseg;
 478         required = nilfs_segctor_segsum_block_required(
 479                 sci, &sci->sc_binfo_ptr, binfo_size);
 480         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 481                 nilfs_segctor_end_finfo(sci, inode);
 482                 err = nilfs_segctor_feed_segment(sci);
 483                 if (err)
 484                         return err;
 485                 goto retry;
 486         }
 487         if (unlikely(required)) {
 488                 err = nilfs_segbuf_extend_segsum(segbuf);
 489                 if (unlikely(err))
 490                         goto failed;
 491         }
 492         if (sci->sc_blk_cnt == 0)
 493                 nilfs_segctor_begin_finfo(sci, inode);
 494
 495         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 496         /* Substitution to vblocknr is delayed until update_blocknr() */
 497         nilfs_segbuf_add_file_buffer(segbuf, bh);
 498         sci->sc_blk_cnt++;
 499  failed:
 500         return err;
 501 }
 502
 503 /*
 504  * Callback functions that enumerate, mark, and collect dirty blocks
 505  */
 506 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 507                                    struct buffer_head *bh, struct inode *inode)
 508 {
 509         int err;
 510
 511         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 512         if (err < 0)
 513                 return err;
 514
 515         err = nilfs_segctor_add_file_block(sci, bh, inode,
 516                                            sizeof(struct nilfs_binfo_v));
 517         if (!err)
 518                 sci->sc_datablk_cnt++;
 519         return err;
 520 }
 521
 522 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 523                                    struct buffer_head *bh,
 524                                    struct inode *inode)
 525 {
 526         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 527 }
 528
 529 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 530                                    struct buffer_head *bh,
 531                                    struct inode *inode)
 532 {
 533         WARN_ON(!buffer_dirty(bh));
 534         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 535 }
 536
 537 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 538                                         struct nilfs_segsum_pointer *ssp,
 539                                         union nilfs_binfo *binfo)
 540 {
 541         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 542                 sci, ssp, sizeof(*binfo_v));
 543         *binfo_v = binfo->bi_v;
 544 }
 545
 546 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 547                                         struct nilfs_segsum_pointer *ssp,
 548                                         union nilfs_binfo *binfo)
 549 {
 550         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 551                 sci, ssp, sizeof(*vblocknr));
 552         *vblocknr = binfo->bi_v.bi_vblocknr;
 553 }
 554
 555 static struct nilfs_sc_operations nilfs_sc_file_ops = {
 556         .collect_data = nilfs_collect_file_data,
 557         .collect_node = nilfs_collect_file_node,
 558         .collect_bmap = nilfs_collect_file_bmap,
 559         .write_data_binfo = nilfs_write_file_data_binfo,
 560         .write_node_binfo = nilfs_write_file_node_binfo,
 561 };
 562
 563 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 564                                   struct buffer_head *bh, struct inode *inode)
 565 {
 566         int err;
 567
 568         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 569         if (err < 0)
 570                 return err;
 571
 572         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 573         if (!err)
 574                 sci->sc_datablk_cnt++;
 575         return err;
 576 }
 577
 578 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 579                                   struct buffer_head *bh, struct inode *inode)
 580 {
 581         WARN_ON(!buffer_dirty(bh));
 582         return nilfs_segctor_add_file_block(sci, bh, inode,
 583                                             sizeof(struct nilfs_binfo_dat));
 584 }
 585
 586 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 587                                        struct nilfs_segsum_pointer *ssp,
 588                                        union nilfs_binfo *binfo)
 589 {
 590         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 591                                                           sizeof(*blkoff));
 592         *blkoff = binfo->bi_dat.bi_blkoff;
 593 }
 594
 595 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 596                                        struct nilfs_segsum_pointer *ssp,
 597                                        union nilfs_binfo *binfo)
 598 {
 599         struct nilfs_binfo_dat *binfo_dat =
 600                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 601         *binfo_dat = binfo->bi_dat;
 602 }
 603
 604 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
 605         .collect_data = nilfs_collect_dat_data,
 606         .collect_node = nilfs_collect_file_node,
 607         .collect_bmap = nilfs_collect_dat_bmap,
 608         .write_data_binfo = nilfs_write_dat_data_binfo,
 609         .write_node_binfo = nilfs_write_dat_node_binfo,
 610 };
 611
 612 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 613         .collect_data = nilfs_collect_file_data,
 614         .collect_node = NULL,
 615         .collect_bmap = NULL,
 616         .write_data_binfo = nilfs_write_file_data_binfo,
 617         .write_node_binfo = NULL,
 618 };
 619
 620 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 621                                               struct list_head *listp,
 622                                               size_t nlimit,
 623                                               loff_t start, loff_t end)
 624 {
 625         struct address_space *mapping = inode->i_mapping;
 626         struct pagevec pvec;
 627         pgoff_t index = 0, last = ULONG_MAX;
 628         size_t ndirties = 0;
 629         int i;
 630
 631         if (unlikely(start != 0 || end != LLONG_MAX)) {
 632                 /*
 633                  * A valid range is given for sync-ing data pages. The
 634                  * range is rounded to per-page; extra dirty buffers
 635                  * may be included if blocksize < pagesize.
 636                  */
 637                 index = start >> PAGE_SHIFT;
 638                 last = end >> PAGE_SHIFT;
 639         }
 640         pagevec_init(&pvec, 0);
 641  repeat:
 642         if (unlikely(index > last) ||
 643             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 644                                 min_t(pgoff_t, last - index,
 645                                       PAGEVEC_SIZE - 1) + 1))
 646                 return ndirties;
 647
 648         for (i = 0; i < pagevec_count(&pvec); i++) {
 649                 struct buffer_head *bh, *head;
 650                 struct page *page = pvec.pages[i];
 651
 652                 if (unlikely(page->index > last))
 653                         break;
 654
 655                 lock_page(page);
 656                 if (!page_has_buffers(page))
 657                         create_empty_buffers(page, 1 << inode->i_blkbits, 0);
 658                 unlock_page(page);
 659
 660                 bh = head = page_buffers(page);
 661                 do {
 662                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 663                                 continue;
 664                         get_bh(bh);
 665                         list_add_tail(&bh->b_assoc_buffers, listp);
 666                         ndirties++;
 667                         if (unlikely(ndirties >= nlimit)) {
 668                                 pagevec_release(&pvec);
 669                                 cond_resched();
 670                                 return ndirties;
 671                         }
 672                 } while (bh = bh->b_this_page, bh != head);
 673         }
 674         pagevec_release(&pvec);
 675         cond_resched();
 676         goto repeat;
 677 }
 678
 679 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 680                                             struct list_head *listp)
 681 {
 682         struct nilfs_inode_info *ii = NILFS_I(inode);
 683         struct address_space *mapping = &ii->i_btnode_cache;
 684         struct pagevec pvec;
 685         struct buffer_head *bh, *head;
 686         unsigned int i;
 687         pgoff_t index = 0;
 688
 689         pagevec_init(&pvec, 0);
 690
 691         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 692                                   PAGEVEC_SIZE)) {
 693                 for (i = 0; i < pagevec_count(&pvec); i++) {
 694                         bh = head = page_buffers(pvec.pages[i]);
 695                         do {
 696                                 if (buffer_dirty(bh) &&
 697                                                 !buffer_async_write(bh)) {
 698                                         get_bh(bh);
 699                                         list_add_tail(&bh->b_assoc_buffers,
 700                                                       listp);
 701                                 }
 702                                 bh = bh->b_this_page;
 703                         } while (bh != head);
 704                 }
 705                 pagevec_release(&pvec);
 706                 cond_resched();
 707         }
 708 }
 709
 710 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 711                                struct list_head *head, int force)
 712 {
 713         struct nilfs_inode_info *ii, *n;
 714         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 715         unsigned nv = 0;
 716
 717         while (!list_empty(head)) {
 718                 spin_lock(&nilfs->ns_inode_lock);
 719                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 720                         list_del_init(&ii->i_dirty);
 721                         if (force) {
 722                                 if (unlikely(ii->i_bh)) {
 723                                         brelse(ii->i_bh);
 724                                         ii->i_bh = NULL;
 725                                 }
 726                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 727                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 728                                 list_add_tail(&ii->i_dirty,
 729                                               &nilfs->ns_dirty_files);
 730                                 continue;
 731                         }
 732                         ivec[nv++] = ii;
 733                         if (nv == SC_N_INODEVEC)
 734                                 break;
 735                 }
 736                 spin_unlock(&nilfs->ns_inode_lock);
 737
 738                 for (pii = ivec; nv > 0; pii++, nv--)
 739                         iput(&(*pii)->vfs_inode);
 740         }
 741 }
 742
 743 static void nilfs_iput_work_func(struct work_struct *work)
 744 {
 745         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 746                                                  sc_iput_work);
 747         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 748
 749         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 750 }
 751
 752 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 753                                      struct nilfs_root *root)
 754 {
 755         int ret = 0;
 756
 757         if (nilfs_mdt_fetch_dirty(root->ifile))
 758                 ret++;
 759         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 760                 ret++;
 761         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 762                 ret++;
 763         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 764                 ret++;
 765         return ret;
 766 }
 767
 768 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 769 {
 770         return list_empty(&sci->sc_dirty_files) &&
 771                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 772                 sci->sc_nfreesegs == 0 &&
 773                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 774 }
 775
 776 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 777 {
 778         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 779         int ret = 0;
 780
 781         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 782                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 783
 784         spin_lock(&nilfs->ns_inode_lock);
 785         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 786                 ret++;
 787
 788         spin_unlock(&nilfs->ns_inode_lock);
 789         return ret;
 790 }
 791
 792 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 793 {
 794         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 795
 796         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 797         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 798         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 799         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 800 }
 801
 802 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 803 {
 804         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 805         struct buffer_head *bh_cp;
 806         struct nilfs_checkpoint *raw_cp;
 807         int err;
 808
 809         /* XXX: this interface will be changed */
 810         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 811                                           &raw_cp, &bh_cp);
 812         if (likely(!err)) {
 813                 /* The following code is duplicated with cpfile.  But, it is
 814                    needed to collect the checkpoint even if it was not newly
 815                    created */
 816                 mark_buffer_dirty(bh_cp);
 817                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 818                 nilfs_cpfile_put_checkpoint(
 819                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 820         } else
 821                 WARN_ON(err == -EINVAL || err == -ENOENT);
 822
 823         return err;
 824 }
 825
 826 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 827 {
 828         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 829         struct buffer_head *bh_cp;
 830         struct nilfs_checkpoint *raw_cp;
 831         int err;
 832
 833         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 834                                           &raw_cp, &bh_cp);
 835         if (unlikely(err)) {
 836                 WARN_ON(err == -EINVAL || err == -ENOENT);
 837                 goto failed_ibh;
 838         }
 839         raw_cp->cp_snapshot_list.ssl_next = 0;
 840         raw_cp->cp_snapshot_list.ssl_prev = 0;
 841         raw_cp->cp_inodes_count =
 842                 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
 843         raw_cp->cp_blocks_count =
 844                 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
 845         raw_cp->cp_nblk_inc =
 846                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 847         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 848         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 849
 850         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 851                 nilfs_checkpoint_clear_minor(raw_cp);
 852         else
 853                 nilfs_checkpoint_set_minor(raw_cp);
 854
 855         nilfs_write_inode_common(sci->sc_root->ifile,
 856                                  &raw_cp->cp_ifile_inode, 1);
 857         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 858         return 0;
 859
 860  failed_ibh:
 861         return err;
 862 }
 863
 864 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 865                                     struct nilfs_inode_info *ii)
 866
 867 {
 868         struct buffer_head *ibh;
 869         struct nilfs_inode *raw_inode;
 870
 871         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 872                 ibh = ii->i_bh;
 873                 BUG_ON(!ibh);
 874                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 875                                                   ibh);
 876                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 877                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 878         }
 879 }
 880
 881 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 882 {
 883         struct nilfs_inode_info *ii;
 884
 885         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 886                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 887                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 888         }
 889 }
 890
 891 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 892                                              struct the_nilfs *nilfs)
 893 {
 894         struct buffer_head *bh_sr;
 895         struct nilfs_super_root *raw_sr;
 896         unsigned isz, srsz;
 897
 898         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 899         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 900         isz = nilfs->ns_inode_size;
 901         srsz = NILFS_SR_BYTES(isz);
 902
 903         raw_sr->sr_bytes = cpu_to_le16(srsz);
 904         raw_sr->sr_nongc_ctime
 905                 = cpu_to_le64(nilfs_doing_gc() ?
 906                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 907         raw_sr->sr_flags = 0;
 908
 909         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 910                                  NILFS_SR_DAT_OFFSET(isz), 1);
 911         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 912                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 913         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 914                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 915         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 916 }
 917
 918 static void nilfs_redirty_inodes(struct list_head *head)
 919 {
 920         struct nilfs_inode_info *ii;
 921
 922         list_for_each_entry(ii, head, i_dirty) {
 923                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 924                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 925         }
 926 }
 927
 928 static void nilfs_drop_collected_inodes(struct list_head *head)
 929 {
 930         struct nilfs_inode_info *ii;
 931
 932         list_for_each_entry(ii, head, i_dirty) {
 933                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 934                         continue;
 935
 936                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 937                 set_bit(NILFS_I_UPDATED, &ii->i_state);
 938         }
 939 }
 940
 941 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
 942                                        struct inode *inode,
 943                                        struct list_head *listp,
 944                                        int (*collect)(struct nilfs_sc_info *,
 945                                                       struct buffer_head *,
 946                                                       struct inode *))
 947 {
 948         struct buffer_head *bh, *n;
 949         int err = 0;
 950
 951         if (collect) {
 952                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
 953                         list_del_init(&bh->b_assoc_buffers);
 954                         err = collect(sci, bh, inode);
 955                         brelse(bh);
 956                         if (unlikely(err))
 957                                 goto dispose_buffers;
 958                 }
 959                 return 0;
 960         }
 961
 962  dispose_buffers:
 963         while (!list_empty(listp)) {
 964                 bh = list_first_entry(listp, struct buffer_head,
 965                                       b_assoc_buffers);
 966                 list_del_init(&bh->b_assoc_buffers);
 967                 brelse(bh);
 968         }
 969         return err;
 970 }
 971
 972 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
 973 {
 974         /* Remaining number of blocks within segment buffer */
 975         return sci->sc_segbuf_nblocks -
 976                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
 977 }
 978
 979 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
 980                                    struct inode *inode,
 981                                    struct nilfs_sc_operations *sc_ops)
 982 {
 983         LIST_HEAD(data_buffers);
 984         LIST_HEAD(node_buffers);
 985         int err;
 986
 987         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
 988                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
 989
 990                 n = nilfs_lookup_dirty_data_buffers(
 991                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
 992                 if (n > rest) {
 993                         err = nilfs_segctor_apply_buffers(
 994                                 sci, inode, &data_buffers,
 995                                 sc_ops->collect_data);
 996                         BUG_ON(!err); /* always receive -E2BIG or true error */
 997                         goto break_or_fail;
 998                 }
 999         }
1000         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1001
1002         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1003                 err = nilfs_segctor_apply_buffers(
1004                         sci, inode, &data_buffers, sc_ops->collect_data);
1005                 if (unlikely(err)) {
1006                         /* dispose node list */
1007                         nilfs_segctor_apply_buffers(
1008                                 sci, inode, &node_buffers, NULL);
1009                         goto break_or_fail;
1010                 }
1011                 sci->sc_stage.flags |= NILFS_CF_NODE;
1012         }
1013         /* Collect node */
1014         err = nilfs_segctor_apply_buffers(
1015                 sci, inode, &node_buffers, sc_ops->collect_node);
1016         if (unlikely(err))
1017                 goto break_or_fail;
1018
1019         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1020         err = nilfs_segctor_apply_buffers(
1021                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1022         if (unlikely(err))
1023                 goto break_or_fail;
1024
1025         nilfs_segctor_end_finfo(sci, inode);
1026         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1027
1028  break_or_fail:
1029         return err;
1030 }
1031
1032 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1033                                          struct inode *inode)
1034 {
1035         LIST_HEAD(data_buffers);
1036         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1037         int err;
1038
1039         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1040                                             sci->sc_dsync_start,
1041                                             sci->sc_dsync_end);
1042
1043         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1044                                           nilfs_collect_file_data);
1045         if (!err) {
1046                 nilfs_segctor_end_finfo(sci, inode);
1047                 BUG_ON(n > rest);
1048                 /* always receive -E2BIG or true error if n > rest */
1049         }
1050         return err;
1051 }
1052
1053 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1054 {
1055         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1056         struct list_head *head;
1057         struct nilfs_inode_info *ii;
1058         size_t ndone;
1059         int err = 0;
1060
1061         switch (sci->sc_stage.scnt) {
1062         case NILFS_ST_INIT:
1063                 /* Pre-processes */
1064                 sci->sc_stage.flags = 0;
1065
1066                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1067                         sci->sc_nblk_inc = 0;
1068                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1069                         if (mode == SC_LSEG_DSYNC) {
1070                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1071                                 goto dsync_mode;
1072                         }
1073                 }
1074
1075                 sci->sc_stage.dirty_file_ptr = NULL;
1076                 sci->sc_stage.gc_inode_ptr = NULL;
1077                 if (mode == SC_FLUSH_DAT) {
1078                         sci->sc_stage.scnt = NILFS_ST_DAT;
1079                         goto dat_stage;
1080                 }
1081                 sci->sc_stage.scnt++;  /* Fall through */
1082         case NILFS_ST_GC:
1083                 if (nilfs_doing_gc()) {
1084                         head = &sci->sc_gc_inodes;
1085                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1086                                                 head, i_dirty);
1087                         list_for_each_entry_continue(ii, head, i_dirty) {
1088                                 err = nilfs_segctor_scan_file(
1089                                         sci, &ii->vfs_inode,
1090                                         &nilfs_sc_file_ops);
1091                                 if (unlikely(err)) {
1092                                         sci->sc_stage.gc_inode_ptr = list_entry(
1093                                                 ii->i_dirty.prev,
1094                                                 struct nilfs_inode_info,
1095                                                 i_dirty);
1096                                         goto break_or_fail;
1097                                 }
1098                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1099                         }
1100                         sci->sc_stage.gc_inode_ptr = NULL;
1101                 }
1102                 sci->sc_stage.scnt++;  /* Fall through */
1103         case NILFS_ST_FILE:
1104                 head = &sci->sc_dirty_files;
1105                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1106                                         i_dirty);
1107                 list_for_each_entry_continue(ii, head, i_dirty) {
1108                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1109
1110                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1111                                                       &nilfs_sc_file_ops);
1112                         if (unlikely(err)) {
1113                                 sci->sc_stage.dirty_file_ptr =
1114                                         list_entry(ii->i_dirty.prev,
1115                                                    struct nilfs_inode_info,
1116                                                    i_dirty);
1117                                 goto break_or_fail;
1118                         }
1119                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1120                         /* XXX: required ? */
1121                 }
1122                 sci->sc_stage.dirty_file_ptr = NULL;
1123                 if (mode == SC_FLUSH_FILE) {
1124                         sci->sc_stage.scnt = NILFS_ST_DONE;
1125                         return 0;
1126                 }
1127                 sci->sc_stage.scnt++;
1128                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1129                 /* Fall through */
1130         case NILFS_ST_IFILE:
1131                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1132                                               &nilfs_sc_file_ops);
1133                 if (unlikely(err))
1134                         break;
1135                 sci->sc_stage.scnt++;
1136                 /* Creating a checkpoint */
1137                 err = nilfs_segctor_create_checkpoint(sci);
1138                 if (unlikely(err))
1139                         break;
1140                 /* Fall through */
1141         case NILFS_ST_CPFILE:
1142                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1143                                               &nilfs_sc_file_ops);
1144                 if (unlikely(err))
1145                         break;
1146                 sci->sc_stage.scnt++;  /* Fall through */
1147         case NILFS_ST_SUFILE:
1148                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1149                                          sci->sc_nfreesegs, &ndone);
1150                 if (unlikely(err)) {
1151                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1152                                                   sci->sc_freesegs, ndone,
1153                                                   NULL);
1154                         break;
1155                 }
1156                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1157
1158                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1159                                               &nilfs_sc_file_ops);
1160                 if (unlikely(err))
1161                         break;
1162                 sci->sc_stage.scnt++;  /* Fall through */
1163         case NILFS_ST_DAT:
1164  dat_stage:
1165                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1166                                               &nilfs_sc_dat_ops);
1167                 if (unlikely(err))
1168                         break;
1169                 if (mode == SC_FLUSH_DAT) {
1170                         sci->sc_stage.scnt = NILFS_ST_DONE;
1171                         return 0;
1172                 }
1173                 sci->sc_stage.scnt++;  /* Fall through */
1174         case NILFS_ST_SR:
1175                 if (mode == SC_LSEG_SR) {
1176                         /* Appending a super root */
1177                         err = nilfs_segctor_add_super_root(sci);
1178                         if (unlikely(err))
1179                                 break;
1180                 }
1181                 /* End of a logical segment */
1182                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1183                 sci->sc_stage.scnt = NILFS_ST_DONE;
1184                 return 0;
1185         case NILFS_ST_DSYNC:
1186  dsync_mode:
1187                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1188                 ii = sci->sc_dsync_inode;
1189                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1190                         break;
1191
1192                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1193                 if (unlikely(err))
1194                         break;
1195                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1196                 sci->sc_stage.scnt = NILFS_ST_DONE;
1197                 return 0;
1198         case NILFS_ST_DONE:
1199                 return 0;
1200         default:
1201                 BUG();
1202         }
1203
1204  break_or_fail:
1205         return err;
1206 }
1207
1208 /**
1209  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1210  * @sci: nilfs_sc_info
1211  * @nilfs: nilfs object
1212  */
1213 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1214                                             struct the_nilfs *nilfs)
1215 {
1216         struct nilfs_segment_buffer *segbuf, *prev;
1217         __u64 nextnum;
1218         int err, alloc = 0;
1219
1220         segbuf = nilfs_segbuf_new(sci->sc_super);
1221         if (unlikely(!segbuf))
1222                 return -ENOMEM;
1223
1224         if (list_empty(&sci->sc_write_logs)) {
1225                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1226                                  nilfs->ns_pseg_offset, nilfs);
1227                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1228                         nilfs_shift_to_next_segment(nilfs);
1229                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1230                 }
1231
1232                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1233                 nextnum = nilfs->ns_nextnum;
1234
1235                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1236                         /* Start from the head of a new full segment */
1237                         alloc++;
1238         } else {
1239                 /* Continue logs */
1240                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1241                 nilfs_segbuf_map_cont(segbuf, prev);
1242                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1243                 nextnum = prev->sb_nextnum;
1244
1245                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1246                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1247                         segbuf->sb_sum.seg_seq++;
1248                         alloc++;
1249                 }
1250         }
1251
1252         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1253         if (err)
1254                 goto failed;
1255
1256         if (alloc) {
1257                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1258                 if (err)
1259                         goto failed;
1260         }
1261         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1262
1263         BUG_ON(!list_empty(&sci->sc_segbufs));
1264         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1265         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1266         return 0;
1267
1268  failed:
1269         nilfs_segbuf_free(segbuf);
1270         return err;
1271 }
1272
1273 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1274                                          struct the_nilfs *nilfs, int nadd)
1275 {
1276         struct nilfs_segment_buffer *segbuf, *prev;
1277         struct inode *sufile = nilfs->ns_sufile;
1278         __u64 nextnextnum;
1279         LIST_HEAD(list);
1280         int err, ret, i;
1281
1282         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1283         /*
1284          * Since the segment specified with nextnum might be allocated during
1285          * the previous construction, the buffer including its segusage may
1286          * not be dirty.  The following call ensures that the buffer is dirty
1287          * and will pin the buffer on memory until the sufile is written.
1288          */
1289         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1290         if (unlikely(err))
1291                 return err;
1292
1293         for (i = 0; i < nadd; i++) {
1294                 /* extend segment info */
1295                 err = -ENOMEM;
1296                 segbuf = nilfs_segbuf_new(sci->sc_super);
1297                 if (unlikely(!segbuf))
1298                         goto failed;
1299
1300                 /* map this buffer to region of segment on-disk */
1301                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1302                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1303
1304                 /* allocate the next next full segment */
1305                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1306                 if (unlikely(err))
1307                         goto failed_segbuf;
1308
1309                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1310                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1311
1312                 list_add_tail(&segbuf->sb_list, &list);
1313                 prev = segbuf;
1314         }
1315         list_splice_tail(&list, &sci->sc_segbufs);
1316         return 0;
1317
1318  failed_segbuf:
1319         nilfs_segbuf_free(segbuf);
1320  failed:
1321         list_for_each_entry(segbuf, &list, sb_list) {
1322                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1323                 WARN_ON(ret); /* never fails */
1324         }
1325         nilfs_destroy_logs(&list);
1326         return err;
1327 }
1328
1329 static void nilfs_free_incomplete_logs(struct list_head *logs,
1330                                        struct the_nilfs *nilfs)
1331 {
1332         struct nilfs_segment_buffer *segbuf, *prev;
1333         struct inode *sufile = nilfs->ns_sufile;
1334         int ret;
1335
1336         segbuf = NILFS_FIRST_SEGBUF(logs);
1337         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1338                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1339                 WARN_ON(ret); /* never fails */
1340         }
1341         if (atomic_read(&segbuf->sb_err)) {
1342                 /* Case 1: The first segment failed */
1343                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1344                         /* Case 1a:  Partial segment appended into an existing
1345                            segment */
1346                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1347                                                 segbuf->sb_fseg_end);
1348                 else /* Case 1b:  New full segment */
1349                         set_nilfs_discontinued(nilfs);
1350         }
1351
1352         prev = segbuf;
1353         list_for_each_entry_continue(segbuf, logs, sb_list) {
1354                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1355                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1356                         WARN_ON(ret); /* never fails */
1357                 }
1358                 if (atomic_read(&segbuf->sb_err) &&
1359                     segbuf->sb_segnum != nilfs->ns_nextnum)
1360                         /* Case 2: extended segment (!= next) failed */
1361                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1362                 prev = segbuf;
1363         }
1364 }
1365
1366 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1367                                           struct inode *sufile)
1368 {
1369         struct nilfs_segment_buffer *segbuf;
1370         unsigned long live_blocks;
1371         int ret;
1372
1373         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1374                 live_blocks = segbuf->sb_sum.nblocks +
1375                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1376                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1377                                                      live_blocks,
1378                                                      sci->sc_seg_ctime);
1379                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1380         }
1381 }
1382
1383 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1384 {
1385         struct nilfs_segment_buffer *segbuf;
1386         int ret;
1387
1388         segbuf = NILFS_FIRST_SEGBUF(logs);
1389         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1390                                              segbuf->sb_pseg_start -
1391                                              segbuf->sb_fseg_start, 0);
1392         WARN_ON(ret); /* always succeed because the segusage is dirty */
1393
1394         list_for_each_entry_continue(segbuf, logs, sb_list) {
1395                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1396                                                      0, 0);
1397                 WARN_ON(ret); /* always succeed */
1398         }
1399 }
1400
1401 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1402                                             struct nilfs_segment_buffer *last,
1403                                             struct inode *sufile)
1404 {
1405         struct nilfs_segment_buffer *segbuf = last;
1406         int ret;
1407
1408         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1409                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1410                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1411                 WARN_ON(ret);
1412         }
1413         nilfs_truncate_logs(&sci->sc_segbufs, last);
1414 }
1415
1416
1417 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1418                                  struct the_nilfs *nilfs, int mode)
1419 {
1420         struct nilfs_cstage prev_stage = sci->sc_stage;
1421         int err, nadd = 1;
1422
1423         /* Collection retry loop */
1424         for (;;) {
1425                 sci->sc_nblk_this_inc = 0;
1426                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1427
1428                 err = nilfs_segctor_reset_segment_buffer(sci);
1429                 if (unlikely(err))
1430                         goto failed;
1431
1432                 err = nilfs_segctor_collect_blocks(sci, mode);
1433                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1434                 if (!err)
1435                         break;
1436
1437                 if (unlikely(err != -E2BIG))
1438                         goto failed;
1439
1440                 /* The current segment is filled up */
1441                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1442                         break;
1443
1444                 nilfs_clear_logs(&sci->sc_segbufs);
1445
1446                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1447                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1448                                                         sci->sc_freesegs,
1449                                                         sci->sc_nfreesegs,
1450                                                         NULL);
1451                         WARN_ON(err); /* do not happen */
1452                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1453                 }
1454
1455                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1456                 if (unlikely(err))
1457                         return err;
1458
1459                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1460                 sci->sc_stage = prev_stage;
1461         }
1462         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1463         return 0;
1464
1465  failed:
1466         return err;
1467 }
1468
1469 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1470                                       struct buffer_head *new_bh)
1471 {
1472         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1473
1474         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1475         /* The caller must release old_bh */
1476 }
1477
1478 static int
1479 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1480                                      struct nilfs_segment_buffer *segbuf,
1481                                      int mode)
1482 {
1483         struct inode *inode = NULL;
1484         sector_t blocknr;
1485         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1486         unsigned long nblocks = 0, ndatablk = 0;
1487         struct nilfs_sc_operations *sc_op = NULL;
1488         struct nilfs_segsum_pointer ssp;
1489         struct nilfs_finfo *finfo = NULL;
1490         union nilfs_binfo binfo;
1491         struct buffer_head *bh, *bh_org;
1492         ino_t ino = 0;
1493         int err = 0;
1494
1495         if (!nfinfo)
1496                 goto out;
1497
1498         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1499         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1500         ssp.offset = sizeof(struct nilfs_segment_summary);
1501
1502         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1503                 if (bh == segbuf->sb_super_root)
1504                         break;
1505                 if (!finfo) {
1506                         finfo = nilfs_segctor_map_segsum_entry(
1507                                 sci, &ssp, sizeof(*finfo));
1508                         ino = le64_to_cpu(finfo->fi_ino);
1509                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1510                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1511
1512                         inode = bh->b_page->mapping->host;
1513
1514                         if (mode == SC_LSEG_DSYNC)
1515                                 sc_op = &nilfs_sc_dsync_ops;
1516                         else if (ino == NILFS_DAT_INO)
1517                                 sc_op = &nilfs_sc_dat_ops;
1518                         else /* file blocks */
1519                                 sc_op = &nilfs_sc_file_ops;
1520                 }
1521                 bh_org = bh;
1522                 get_bh(bh_org);
1523                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1524                                         &binfo);
1525                 if (bh != bh_org)
1526                         nilfs_list_replace_buffer(bh_org, bh);
1527                 brelse(bh_org);
1528                 if (unlikely(err))
1529                         goto failed_bmap;
1530
1531                 if (ndatablk > 0)
1532                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1533                 else
1534                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1535
1536                 blocknr++;
1537                 if (--nblocks == 0) {
1538                         finfo = NULL;
1539                         if (--nfinfo == 0)
1540                                 break;
1541                 } else if (ndatablk > 0)
1542                         ndatablk--;
1543         }
1544  out:
1545         return 0;
1546
1547  failed_bmap:
1548         return err;
1549 }
1550
1551 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1552 {
1553         struct nilfs_segment_buffer *segbuf;
1554         int err;
1555
1556         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1557                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1558                 if (unlikely(err))
1559                         return err;
1560                 nilfs_segbuf_fill_in_segsum(segbuf);
1561         }
1562         return 0;
1563 }
1564
1565 static void nilfs_begin_page_io(struct page *page)
1566 {
1567         if (!page || PageWriteback(page))
1568                 /* For split b-tree node pages, this function may be called
1569                    twice.  We ignore the 2nd or later calls by this check. */
1570                 return;
1571
1572         lock_page(page);
1573         clear_page_dirty_for_io(page);
1574         set_page_writeback(page);
1575         unlock_page(page);
1576 }
1577
1578 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1579 {
1580         struct nilfs_segment_buffer *segbuf;
1581         struct page *bd_page = NULL, *fs_page = NULL;
1582
1583         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1584                 struct buffer_head *bh;
1585
1586                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1587                                     b_assoc_buffers) {
1588                         set_buffer_async_write(bh);
1589                         if (bh->b_page != bd_page) {
1590                                 if (bd_page) {
1591                                         lock_page(bd_page);
1592                                         clear_page_dirty_for_io(bd_page);
1593                                         set_page_writeback(bd_page);
1594                                         unlock_page(bd_page);
1595                                 }
1596                                 bd_page = bh->b_page;
1597                         }
1598                 }
1599
1600                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1601                                     b_assoc_buffers) {
1602                         set_buffer_async_write(bh);
1603                         if (bh == segbuf->sb_super_root) {
1604                                 if (bh->b_page != bd_page) {
1605                                         lock_page(bd_page);
1606                                         clear_page_dirty_for_io(bd_page);
1607                                         set_page_writeback(bd_page);
1608                                         unlock_page(bd_page);
1609                                         bd_page = bh->b_page;
1610                                 }
1611                                 break;
1612                         }
1613                         if (bh->b_page != fs_page) {
1614                                 nilfs_begin_page_io(fs_page);
1615                                 fs_page = bh->b_page;
1616                         }
1617                 }
1618         }
1619         if (bd_page) {
1620                 lock_page(bd_page);
1621                 clear_page_dirty_for_io(bd_page);
1622                 set_page_writeback(bd_page);
1623                 unlock_page(bd_page);
1624         }
1625         nilfs_begin_page_io(fs_page);
1626 }
1627
1628 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1629                                struct the_nilfs *nilfs)
1630 {
1631         int ret;
1632
1633         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1634         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1635         return ret;
1636 }
1637
1638 static void nilfs_end_page_io(struct page *page, int err)
1639 {
1640         if (!page)
1641                 return;
1642
1643         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1644                 /*
1645                  * For b-tree node pages, this function may be called twice
1646                  * or more because they might be split in a segment.
1647                  */
1648                 if (PageDirty(page)) {
1649                         /*
1650                          * For pages holding split b-tree node buffers, dirty
1651                          * flag on the buffers may be cleared discretely.
1652                          * In that case, the page is once redirtied for
1653                          * remaining buffers, and it must be cancelled if
1654                          * all the buffers get cleaned later.
1655                          */
1656                         lock_page(page);
1657                         if (nilfs_page_buffers_clean(page))
1658                                 __nilfs_clear_page_dirty(page);
1659                         unlock_page(page);
1660                 }
1661                 return;
1662         }
1663
1664         if (!err) {
1665                 if (!nilfs_page_buffers_clean(page))
1666                         __set_page_dirty_nobuffers(page);
1667                 ClearPageError(page);
1668         } else {
1669                 __set_page_dirty_nobuffers(page);
1670                 SetPageError(page);
1671         }
1672
1673         end_page_writeback(page);
1674 }
1675
1676 static void nilfs_abort_logs(struct list_head *logs, int err)
1677 {
1678         struct nilfs_segment_buffer *segbuf;
1679         struct page *bd_page = NULL, *fs_page = NULL;
1680         struct buffer_head *bh;
1681
1682         if (list_empty(logs))
1683                 return;
1684
1685         list_for_each_entry(segbuf, logs, sb_list) {
1686                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1687                                     b_assoc_buffers) {
1688                         clear_buffer_async_write(bh);
1689                         if (bh->b_page != bd_page) {
1690                                 if (bd_page)
1691                                         end_page_writeback(bd_page);
1692                                 bd_page = bh->b_page;
1693                         }
1694                 }
1695
1696                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1697                                     b_assoc_buffers) {
1698                         clear_buffer_async_write(bh);
1699                         if (bh == segbuf->sb_super_root) {
1700                                 if (bh->b_page != bd_page) {
1701                                         end_page_writeback(bd_page);
1702                                         bd_page = bh->b_page;
1703                                 }
1704                                 break;
1705                         }
1706                         if (bh->b_page != fs_page) {
1707                                 nilfs_end_page_io(fs_page, err);
1708                                 fs_page = bh->b_page;
1709                         }
1710                 }
1711         }
1712         if (bd_page)
1713                 end_page_writeback(bd_page);
1714
1715         nilfs_end_page_io(fs_page, err);
1716 }
1717
1718 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1719                                              struct the_nilfs *nilfs, int err)
1720 {
1721         LIST_HEAD(logs);
1722         int ret;
1723
1724         list_splice_tail_init(&sci->sc_write_logs, &logs);
1725         ret = nilfs_wait_on_logs(&logs);
1726         nilfs_abort_logs(&logs, ret ? : err);
1727
1728         list_splice_tail_init(&sci->sc_segbufs, &logs);
1729         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1730         nilfs_free_incomplete_logs(&logs, nilfs);
1731
1732         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1733                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1734                                                 sci->sc_freesegs,
1735                                                 sci->sc_nfreesegs,
1736                                                 NULL);
1737                 WARN_ON(ret); /* do not happen */
1738         }
1739
1740         nilfs_destroy_logs(&logs);
1741 }
1742
1743 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1744                                    struct nilfs_segment_buffer *segbuf)
1745 {
1746         nilfs->ns_segnum = segbuf->sb_segnum;
1747         nilfs->ns_nextnum = segbuf->sb_nextnum;
1748         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1749                 + segbuf->sb_sum.nblocks;
1750         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1751         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1752 }
1753
1754 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1755 {
1756         struct nilfs_segment_buffer *segbuf;
1757         struct page *bd_page = NULL, *fs_page = NULL;
1758         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1759         int update_sr = false;
1760
1761         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1762                 struct buffer_head *bh;
1763
1764                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1765                                     b_assoc_buffers) {
1766                         set_buffer_uptodate(bh);
1767                         clear_buffer_dirty(bh);
1768                         clear_buffer_async_write(bh);
1769                         if (bh->b_page != bd_page) {
1770                                 if (bd_page)
1771                                         end_page_writeback(bd_page);
1772                                 bd_page = bh->b_page;
1773                         }
1774                 }
1775                 /*
1776                  * We assume that the buffers which belong to the same page
1777                  * continue over the buffer list.
1778                  * Under this assumption, the last BHs of pages is
1779                  * identifiable by the discontinuity of bh->b_page
1780                  * (page != fs_page).
1781                  *
1782                  * For B-tree node blocks, however, this assumption is not
1783                  * guaranteed.  The cleanup code of B-tree node pages needs
1784                  * special care.
1785                  */
1786                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1787                                     b_assoc_buffers) {
1788                         set_buffer_uptodate(bh);
1789                         clear_buffer_dirty(bh);
1790                         clear_buffer_async_write(bh);
1791                         clear_buffer_delay(bh);
1792                         clear_buffer_nilfs_volatile(bh);
1793                         clear_buffer_nilfs_redirected(bh);
1794                         if (bh == segbuf->sb_super_root) {
1795                                 if (bh->b_page != bd_page) {
1796                                         end_page_writeback(bd_page);
1797                                         bd_page = bh->b_page;
1798                                 }
1799                                 update_sr = true;
1800                                 break;
1801                         }
1802                         if (bh->b_page != fs_page) {
1803                                 nilfs_end_page_io(fs_page, 0);
1804                                 fs_page = bh->b_page;
1805                         }
1806                 }
1807
1808                 if (!nilfs_segbuf_simplex(segbuf)) {
1809                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1810                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1811                                 sci->sc_lseg_stime = jiffies;
1812                         }
1813                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1814                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1815                 }
1816         }
1817         /*
1818          * Since pages may continue over multiple segment buffers,
1819          * end of the last page must be checked outside of the loop.
1820          */
1821         if (bd_page)
1822                 end_page_writeback(bd_page);
1823
1824         nilfs_end_page_io(fs_page, 0);
1825
1826         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1827
1828         if (nilfs_doing_gc())
1829                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1830         else
1831                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1832
1833         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1834
1835         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1836         nilfs_set_next_segment(nilfs, segbuf);
1837
1838         if (update_sr) {
1839                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1840                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1841
1842                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1843                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1844                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1845                 nilfs_segctor_clear_metadata_dirty(sci);
1846         } else
1847                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1848 }
1849
1850 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1851 {
1852         int ret;
1853
1854         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1855         if (!ret) {
1856                 nilfs_segctor_complete_write(sci);
1857                 nilfs_destroy_logs(&sci->sc_write_logs);
1858         }
1859         return ret;
1860 }
1861
1862 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1863                                              struct the_nilfs *nilfs)
1864 {
1865         struct nilfs_inode_info *ii, *n;
1866         struct inode *ifile = sci->sc_root->ifile;
1867
1868         spin_lock(&nilfs->ns_inode_lock);
1869  retry:
1870         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1871                 if (!ii->i_bh) {
1872                         struct buffer_head *ibh;
1873                         int err;
1874
1875                         spin_unlock(&nilfs->ns_inode_lock);
1876                         err = nilfs_ifile_get_inode_block(
1877                                 ifile, ii->vfs_inode.i_ino, &ibh);
1878                         if (unlikely(err)) {
1879                                 nilfs_warning(sci->sc_super, __func__,
1880                                               "failed to get inode block.\n");
1881                                 return err;
1882                         }
1883                         mark_buffer_dirty(ibh);
1884                         nilfs_mdt_mark_dirty(ifile);
1885                         spin_lock(&nilfs->ns_inode_lock);
1886                         if (likely(!ii->i_bh))
1887                                 ii->i_bh = ibh;
1888                         else
1889                                 brelse(ibh);
1890                         goto retry;
1891                 }
1892
1893                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1894                 set_bit(NILFS_I_BUSY, &ii->i_state);
1895                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1896         }
1897         spin_unlock(&nilfs->ns_inode_lock);
1898
1899         return 0;
1900 }
1901
1902 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1903                                              struct the_nilfs *nilfs)
1904 {
1905         struct nilfs_inode_info *ii, *n;
1906         int during_mount = !(sci->sc_super->s_flags & MS_ACTIVE);
1907         int defer_iput = false;
1908
1909         spin_lock(&nilfs->ns_inode_lock);
1910         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1911                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1912                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1913                         continue;
1914
1915                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1916                 brelse(ii->i_bh);
1917                 ii->i_bh = NULL;
1918                 list_del_init(&ii->i_dirty);
1919                 if (!ii->vfs_inode.i_nlink || during_mount) {
1920                         /*
1921                          * Defer calling iput() to avoid deadlocks if
1922                          * i_nlink == 0 or mount is not yet finished.
1923                          */
1924                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
1925                         defer_iput = true;
1926                 } else {
1927                         spin_unlock(&nilfs->ns_inode_lock);
1928                         iput(&ii->vfs_inode);
1929                         spin_lock(&nilfs->ns_inode_lock);
1930                 }
1931         }
1932         spin_unlock(&nilfs->ns_inode_lock);
1933
1934         if (defer_iput)
1935                 schedule_work(&sci->sc_iput_work);
1936 }
1937
1938 /*
1939  * Main procedure of segment constructor
1940  */
1941 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
1942 {
1943         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1944         int err;
1945
1946         sci->sc_stage.scnt = NILFS_ST_INIT;
1947         sci->sc_cno = nilfs->ns_cno;
1948
1949         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
1950         if (unlikely(err))
1951                 goto out;
1952
1953         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
1954                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1955
1956         if (nilfs_segctor_clean(sci))
1957                 goto out;
1958
1959         do {
1960                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
1961
1962                 err = nilfs_segctor_begin_construction(sci, nilfs);
1963                 if (unlikely(err))
1964                         goto out;
1965
1966                 /* Update time stamp */
1967                 sci->sc_seg_ctime = get_seconds();
1968
1969                 err = nilfs_segctor_collect(sci, nilfs, mode);
1970                 if (unlikely(err))
1971                         goto failed;
1972
1973                 /* Avoid empty segment */
1974                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
1975                     nilfs_segbuf_empty(sci->sc_curseg)) {
1976                         nilfs_segctor_abort_construction(sci, nilfs, 1);
1977                         goto out;
1978                 }
1979
1980                 err = nilfs_segctor_assign(sci, mode);
1981                 if (unlikely(err))
1982                         goto failed;
1983
1984                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
1985                         nilfs_segctor_fill_in_file_bmap(sci);
1986
1987                 if (mode == SC_LSEG_SR &&
1988                     sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
1989                         err = nilfs_segctor_fill_in_checkpoint(sci);
1990                         if (unlikely(err))
1991                                 goto failed_to_write;
1992
1993                         nilfs_segctor_fill_in_super_root(sci, nilfs);
1994                 }
1995                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
1996
1997                 /* Write partial segments */
1998                 nilfs_segctor_prepare_write(sci);
1999
2000                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2001                                             nilfs->ns_crc_seed);
2002
2003                 err = nilfs_segctor_write(sci, nilfs);
2004                 if (unlikely(err))
2005                         goto failed_to_write;
2006
2007                 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2008                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2009                         /*
2010                          * At this point, we avoid double buffering
2011                          * for blocksize < pagesize because page dirty
2012                          * flag is turned off during write and dirty
2013                          * buffers are not properly collected for
2014                          * pages crossing over segments.
2015                          */
2016                         err = nilfs_segctor_wait(sci);
2017                         if (err)
2018                                 goto failed_to_write;
2019                 }
2020         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2021
2022  out:
2023         nilfs_segctor_drop_written_files(sci, nilfs);
2024         return err;
2025
2026  failed_to_write:
2027         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2028                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2029
2030  failed:
2031         if (nilfs_doing_gc())
2032                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2033         nilfs_segctor_abort_construction(sci, nilfs, err);
2034         goto out;
2035 }
2036
2037 /**
2038  * nilfs_segctor_start_timer - set timer of background write
2039  * @sci: nilfs_sc_info
2040  *
2041  * If the timer has already been set, it ignores the new request.
2042  * This function MUST be called within a section locking the segment
2043  * semaphore.
2044  */
2045 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2046 {
2047         spin_lock(&sci->sc_state_lock);
2048         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2049                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2050                 add_timer(&sci->sc_timer);
2051                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2052         }
2053         spin_unlock(&sci->sc_state_lock);
2054 }
2055
2056 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2057 {
2058         spin_lock(&sci->sc_state_lock);
2059         if (!(sci->sc_flush_request & (1 << bn))) {
2060                 unsigned long prev_req = sci->sc_flush_request;
2061
2062                 sci->sc_flush_request |= (1 << bn);
2063                 if (!prev_req)
2064                         wake_up(&sci->sc_wait_daemon);
2065         }
2066         spin_unlock(&sci->sc_state_lock);
2067 }
2068
2069 /**
2070  * nilfs_flush_segment - trigger a segment construction for resource control
2071  * @sb: super block
2072  * @ino: inode number of the file to be flushed out.
2073  */
2074 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2075 {
2076         struct the_nilfs *nilfs = sb->s_fs_info;
2077         struct nilfs_sc_info *sci = nilfs->ns_writer;
2078
2079         if (!sci || nilfs_doing_construction())
2080                 return;
2081         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2082                                         /* assign bit 0 to data files */
2083 }
2084
2085 struct nilfs_segctor_wait_request {
2086         wait_queue_t    wq;
2087         __u32           seq;
2088         int             err;
2089         atomic_t        done;
2090 };
2091
2092 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2093 {
2094         struct nilfs_segctor_wait_request wait_req;
2095         int err = 0;
2096
2097         spin_lock(&sci->sc_state_lock);
2098         init_wait(&wait_req.wq);
2099         wait_req.err = 0;
2100         atomic_set(&wait_req.done, 0);
2101         wait_req.seq = ++sci->sc_seq_request;
2102         spin_unlock(&sci->sc_state_lock);
2103
2104         init_waitqueue_entry(&wait_req.wq, current);
2105         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2106         set_current_state(TASK_INTERRUPTIBLE);
2107         wake_up(&sci->sc_wait_daemon);
2108
2109         for (;;) {
2110                 if (atomic_read(&wait_req.done)) {
2111                         err = wait_req.err;
2112                         break;
2113                 }
2114                 if (!signal_pending(current)) {
2115                         schedule();
2116                         continue;
2117                 }
2118                 err = -ERESTARTSYS;
2119                 break;
2120         }
2121         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2122         return err;
2123 }
2124
2125 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2126 {
2127         struct nilfs_segctor_wait_request *wrq, *n;
2128         unsigned long flags;
2129
2130         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2131         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2132                                  wq.task_list) {
2133                 if (!atomic_read(&wrq->done) &&
2134                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2135                         wrq->err = err;
2136                         atomic_set(&wrq->done, 1);
2137                 }
2138                 if (atomic_read(&wrq->done)) {
2139                         wrq->wq.func(&wrq->wq,
2140                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2141                                      0, NULL);
2142                 }
2143         }
2144         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2145 }
2146
2147 /**
2148  * nilfs_construct_segment - construct a logical segment
2149  * @sb: super block
2150  *
2151  * Return Value: On success, 0 is retured. On errors, one of the following
2152  * negative error code is returned.
2153  *
2154  * %-EROFS - Read only filesystem.
2155  *
2156  * %-EIO - I/O error
2157  *
2158  * %-ENOSPC - No space left on device (only in a panic state).
2159  *
2160  * %-ERESTARTSYS - Interrupted.
2161  *
2162  * %-ENOMEM - Insufficient memory available.
2163  */
2164 int nilfs_construct_segment(struct super_block *sb)
2165 {
2166         struct the_nilfs *nilfs = sb->s_fs_info;
2167         struct nilfs_sc_info *sci = nilfs->ns_writer;
2168         struct nilfs_transaction_info *ti;
2169         int err;
2170
2171         if (!sci)
2172                 return -EROFS;
2173
2174         /* A call inside transactions causes a deadlock. */
2175         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2176
2177         err = nilfs_segctor_sync(sci);
2178         return err;
2179 }
2180
2181 /**
2182  * nilfs_construct_dsync_segment - construct a data-only logical segment
2183  * @sb: super block
2184  * @inode: inode whose data blocks should be written out
2185  * @start: start byte offset
2186  * @end: end byte offset (inclusive)
2187  *
2188  * Return Value: On success, 0 is retured. On errors, one of the following
2189  * negative error code is returned.
2190  *
2191  * %-EROFS - Read only filesystem.
2192  *
2193  * %-EIO - I/O error
2194  *
2195  * %-ENOSPC - No space left on device (only in a panic state).
2196  *
2197  * %-ERESTARTSYS - Interrupted.
2198  *
2199  * %-ENOMEM - Insufficient memory available.
2200  */
2201 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2202                                   loff_t start, loff_t end)
2203 {
2204         struct the_nilfs *nilfs = sb->s_fs_info;
2205         struct nilfs_sc_info *sci = nilfs->ns_writer;
2206         struct nilfs_inode_info *ii;
2207         struct nilfs_transaction_info ti;
2208         int err = 0;
2209
2210         if (!sci)
2211                 return -EROFS;
2212
2213         nilfs_transaction_lock(sb, &ti, 0);
2214
2215         ii = NILFS_I(inode);
2216         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2217             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2218             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2219             nilfs_discontinued(nilfs)) {
2220                 nilfs_transaction_unlock(sb);
2221                 err = nilfs_segctor_sync(sci);
2222                 return err;
2223         }
2224
2225         spin_lock(&nilfs->ns_inode_lock);
2226         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2227             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2228                 spin_unlock(&nilfs->ns_inode_lock);
2229                 nilfs_transaction_unlock(sb);
2230                 return 0;
2231         }
2232         spin_unlock(&nilfs->ns_inode_lock);
2233         sci->sc_dsync_inode = ii;
2234         sci->sc_dsync_start = start;
2235         sci->sc_dsync_end = end;
2236
2237         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2238
2239         nilfs_transaction_unlock(sb);
2240         return err;
2241 }
2242
2243 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2244 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2245
2246 /**
2247  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2248  * @sci: segment constructor object
2249  */
2250 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2251 {
2252         spin_lock(&sci->sc_state_lock);
2253         sci->sc_seq_accepted = sci->sc_seq_request;
2254         spin_unlock(&sci->sc_state_lock);
2255         del_timer_sync(&sci->sc_timer);
2256 }
2257
2258 /**
2259  * nilfs_segctor_notify - notify the result of request to caller threads
2260  * @sci: segment constructor object
2261  * @mode: mode of log forming
2262  * @err: error code to be notified
2263  */
2264 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2265 {
2266         /* Clear requests (even when the construction failed) */
2267         spin_lock(&sci->sc_state_lock);
2268
2269         if (mode == SC_LSEG_SR) {
2270                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2271                 sci->sc_seq_done = sci->sc_seq_accepted;
2272                 nilfs_segctor_wakeup(sci, err);
2273                 sci->sc_flush_request = 0;
2274         } else {
2275                 if (mode == SC_FLUSH_FILE)
2276                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2277                 else if (mode == SC_FLUSH_DAT)
2278                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2279
2280                 /* re-enable timer if checkpoint creation was not done */
2281                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2282                     time_before(jiffies, sci->sc_timer.expires))
2283                         add_timer(&sci->sc_timer);
2284         }
2285         spin_unlock(&sci->sc_state_lock);
2286 }
2287
2288 /**
2289  * nilfs_segctor_construct - form logs and write them to disk
2290  * @sci: segment constructor object
2291  * @mode: mode of log forming
2292  */
2293 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2294 {
2295         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2296         struct nilfs_super_block **sbp;
2297         int err = 0;
2298
2299         nilfs_segctor_accept(sci);
2300
2301         if (nilfs_discontinued(nilfs))
2302                 mode = SC_LSEG_SR;
2303         if (!nilfs_segctor_confirm(sci))
2304                 err = nilfs_segctor_do_construct(sci, mode);
2305
2306         if (likely(!err)) {
2307                 if (mode != SC_FLUSH_DAT)
2308                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2309                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2310                     nilfs_discontinued(nilfs)) {
2311                         down_write(&nilfs->ns_sem);
2312                         err = -EIO;
2313                         sbp = nilfs_prepare_super(sci->sc_super,
2314                                                   nilfs_sb_will_flip(nilfs));
2315                         if (likely(sbp)) {
2316                                 nilfs_set_log_cursor(sbp[0], nilfs);
2317                                 err = nilfs_commit_super(sci->sc_super,
2318                                                          NILFS_SB_COMMIT);
2319                         }
2320                         up_write(&nilfs->ns_sem);
2321                 }
2322         }
2323
2324         nilfs_segctor_notify(sci, mode, err);
2325         return err;
2326 }
2327
2328 static void nilfs_construction_timeout(unsigned long data)
2329 {
2330         struct task_struct *p = (struct task_struct *)data;
2331         wake_up_process(p);
2332 }
2333
2334 static void
2335 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2336 {
2337         struct nilfs_inode_info *ii, *n;
2338
2339         list_for_each_entry_safe(ii, n, head, i_dirty) {
2340                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2341                         continue;
2342                 list_del_init(&ii->i_dirty);
2343                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2344                 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2345                 iput(&ii->vfs_inode);
2346         }
2347 }
2348
2349 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2350                          void **kbufs)
2351 {
2352         struct the_nilfs *nilfs = sb->s_fs_info;
2353         struct nilfs_sc_info *sci = nilfs->ns_writer;
2354         struct nilfs_transaction_info ti;
2355         int err;
2356
2357         if (unlikely(!sci))
2358                 return -EROFS;
2359
2360         nilfs_transaction_lock(sb, &ti, 1);
2361
2362         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2363         if (unlikely(err))
2364                 goto out_unlock;
2365
2366         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2367         if (unlikely(err)) {
2368                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2369                 goto out_unlock;
2370         }
2371
2372         sci->sc_freesegs = kbufs[4];
2373         sci->sc_nfreesegs = argv[4].v_nmembs;
2374         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2375
2376         for (;;) {
2377                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2378                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2379
2380                 if (likely(!err))
2381                         break;
2382
2383                 nilfs_warning(sb, __func__,
2384                               "segment construction failed. (err=%d)", err);
2385                 set_current_state(TASK_INTERRUPTIBLE);
2386                 schedule_timeout(sci->sc_interval);
2387         }
2388         if (nilfs_test_opt(nilfs, DISCARD)) {
2389                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2390                                                  sci->sc_nfreesegs);
2391                 if (ret) {
2392                         printk(KERN_WARNING
2393                                "NILFS warning: error %d on discard request, "
2394                                "turning discards off for the device\n", ret);
2395                         nilfs_clear_opt(nilfs, DISCARD);
2396                 }
2397         }
2398
2399  out_unlock:
2400         sci->sc_freesegs = NULL;
2401         sci->sc_nfreesegs = 0;
2402         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2403         nilfs_transaction_unlock(sb);
2404         return err;
2405 }
2406
2407 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2408 {
2409         struct nilfs_transaction_info ti;
2410
2411         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2412         nilfs_segctor_construct(sci, mode);
2413
2414         /*
2415          * Unclosed segment should be retried.  We do this using sc_timer.
2416          * Timeout of sc_timer will invoke complete construction which leads
2417          * to close the current logical segment.
2418          */
2419         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2420                 nilfs_segctor_start_timer(sci);
2421
2422         nilfs_transaction_unlock(sci->sc_super);
2423 }
2424
2425 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2426 {
2427         int mode = 0;
2428         int err;
2429
2430         spin_lock(&sci->sc_state_lock);
2431         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2432                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2433         spin_unlock(&sci->sc_state_lock);
2434
2435         if (mode) {
2436                 err = nilfs_segctor_do_construct(sci, mode);
2437
2438                 spin_lock(&sci->sc_state_lock);
2439                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2440                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2441                 spin_unlock(&sci->sc_state_lock);
2442         }
2443         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2444 }
2445
2446 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2447 {
2448         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2449             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2450                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2451                         return SC_FLUSH_FILE;
2452                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2453                         return SC_FLUSH_DAT;
2454         }
2455         return SC_LSEG_SR;
2456 }
2457
2458 /**
2459  * nilfs_segctor_thread - main loop of the segment constructor thread.
2460  * @arg: pointer to a struct nilfs_sc_info.
2461  *
2462  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2463  * to execute segment constructions.
2464  */
2465 static int nilfs_segctor_thread(void *arg)
2466 {
2467         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2468         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2469         int timeout = 0;
2470
2471         sci->sc_timer.data = (unsigned long)current;
2472         sci->sc_timer.function = nilfs_construction_timeout;
2473
2474         /* start sync. */
2475         sci->sc_task = current;
2476         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2477         printk(KERN_INFO
2478                "segctord starting. Construction interval = %lu seconds, "
2479                "CP frequency < %lu seconds\n",
2480                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2481
2482         spin_lock(&sci->sc_state_lock);
2483  loop:
2484         for (;;) {
2485                 int mode;
2486
2487                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2488                         goto end_thread;
2489
2490                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2491                         mode = SC_LSEG_SR;
2492                 else if (!sci->sc_flush_request)
2493                         break;
2494                 else
2495                         mode = nilfs_segctor_flush_mode(sci);
2496
2497                 spin_unlock(&sci->sc_state_lock);
2498                 nilfs_segctor_thread_construct(sci, mode);
2499                 spin_lock(&sci->sc_state_lock);
2500                 timeout = 0;
2501         }
2502
2503
2504         if (freezing(current)) {
2505                 spin_unlock(&sci->sc_state_lock);
2506                 refrigerator();
2507                 spin_lock(&sci->sc_state_lock);
2508         } else {
2509                 DEFINE_WAIT(wait);
2510                 int should_sleep = 1;
2511
2512                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2513                                 TASK_INTERRUPTIBLE);
2514
2515                 if (sci->sc_seq_request != sci->sc_seq_done)
2516                         should_sleep = 0;
2517                 else if (sci->sc_flush_request)
2518                         should_sleep = 0;
2519                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2520                         should_sleep = time_before(jiffies,
2521                                         sci->sc_timer.expires);
2522
2523                 if (should_sleep) {
2524                         spin_unlock(&sci->sc_state_lock);
2525                         schedule();
2526                         spin_lock(&sci->sc_state_lock);
2527                 }
2528                 finish_wait(&sci->sc_wait_daemon, &wait);
2529                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2530                            time_after_eq(jiffies, sci->sc_timer.expires));
2531
2532                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2533                         set_nilfs_discontinued(nilfs);
2534         }
2535         goto loop;
2536
2537  end_thread:
2538         spin_unlock(&sci->sc_state_lock);
2539
2540         /* end sync. */
2541         sci->sc_task = NULL;
2542         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2543         return 0;
2544 }
2545
2546 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2547 {
2548         struct task_struct *t;
2549
2550         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2551         if (IS_ERR(t)) {
2552                 int err = PTR_ERR(t);
2553
2554                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2555                        err);
2556                 return err;
2557         }
2558         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2559         return 0;
2560 }
2561
2562 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2563         __acquires(&sci->sc_state_lock)
2564         __releases(&sci->sc_state_lock)
2565 {
2566         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2567
2568         while (sci->sc_task) {
2569                 wake_up(&sci->sc_wait_daemon);
2570                 spin_unlock(&sci->sc_state_lock);
2571                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2572                 spin_lock(&sci->sc_state_lock);
2573         }
2574 }
2575
2576 /*
2577  * Setup & clean-up functions
2578  */
2579 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2580                                                struct nilfs_root *root)
2581 {
2582         struct the_nilfs *nilfs = sb->s_fs_info;
2583         struct nilfs_sc_info *sci;
2584
2585         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2586         if (!sci)
2587                 return NULL;
2588
2589         sci->sc_super = sb;
2590
2591         nilfs_get_root(root);
2592         sci->sc_root = root;
2593
2594         init_waitqueue_head(&sci->sc_wait_request);
2595         init_waitqueue_head(&sci->sc_wait_daemon);
2596         init_waitqueue_head(&sci->sc_wait_task);
2597         spin_lock_init(&sci->sc_state_lock);
2598         INIT_LIST_HEAD(&sci->sc_dirty_files);
2599         INIT_LIST_HEAD(&sci->sc_segbufs);
2600         INIT_LIST_HEAD(&sci->sc_write_logs);
2601         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2602         INIT_LIST_HEAD(&sci->sc_iput_queue);
2603         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2604         init_timer(&sci->sc_timer);
2605
2606         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2607         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2608         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2609
2610         if (nilfs->ns_interval)
2611                 sci->sc_interval = HZ * nilfs->ns_interval;
2612         if (nilfs->ns_watermark)
2613                 sci->sc_watermark = nilfs->ns_watermark;
2614         return sci;
2615 }
2616
2617 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2618 {
2619         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2620
2621         /* The segctord thread was stopped and its timer was removed.
2622            But some tasks remain. */
2623         do {
2624                 struct nilfs_transaction_info ti;
2625
2626                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2627                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2628                 nilfs_transaction_unlock(sci->sc_super);
2629
2630                 flush_work(&sci->sc_iput_work);
2631
2632         } while (ret && retrycount-- > 0);
2633 }
2634
2635 /**
2636  * nilfs_segctor_destroy - destroy the segment constructor.
2637  * @sci: nilfs_sc_info
2638  *
2639  * nilfs_segctor_destroy() kills the segctord thread and frees
2640  * the nilfs_sc_info struct.
2641  * Caller must hold the segment semaphore.
2642  */
2643 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2644 {
2645         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2646         int flag;
2647
2648         up_write(&nilfs->ns_segctor_sem);
2649
2650         spin_lock(&sci->sc_state_lock);
2651         nilfs_segctor_kill_thread(sci);
2652         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2653                 || sci->sc_seq_request != sci->sc_seq_done);
2654         spin_unlock(&sci->sc_state_lock);
2655
2656         if (flush_work(&sci->sc_iput_work))
2657                 flag = true;
2658
2659         if (flag || !nilfs_segctor_confirm(sci))
2660                 nilfs_segctor_write_out(sci);
2661
2662         if (!list_empty(&sci->sc_dirty_files)) {
2663                 nilfs_warning(sci->sc_super, __func__,
2664                               "dirty file(s) after the final construction\n");
2665                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2666         }
2667
2668         if (!list_empty(&sci->sc_iput_queue)) {
2669                 nilfs_warning(sci->sc_super, __func__,
2670                               "iput queue is not empty\n");
2671                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2672         }
2673
2674         WARN_ON(!list_empty(&sci->sc_segbufs));
2675         WARN_ON(!list_empty(&sci->sc_write_logs));
2676
2677         nilfs_put_root(sci->sc_root);
2678
2679         down_write(&nilfs->ns_segctor_sem);
2680
2681         del_timer_sync(&sci->sc_timer);
2682         kfree(sci);
2683 }
2684
2685 /**
2686  * nilfs_attach_log_writer - attach log writer
2687  * @sb: super block instance
2688  * @root: root object of the current filesystem tree
2689  *
2690  * This allocates a log writer object, initializes it, and starts the
2691  * log writer.
2692  *
2693  * Return Value: On success, 0 is returned. On error, one of the following
2694  * negative error code is returned.
2695  *
2696  * %-ENOMEM - Insufficient memory available.
2697  */
2698 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2699 {
2700         struct the_nilfs *nilfs = sb->s_fs_info;
2701         int err;
2702
2703         if (nilfs->ns_writer) {
2704                 /*
2705                  * This happens if the filesystem was remounted
2706                  * read/write after nilfs_error degenerated it into a
2707                  * read-only mount.
2708                  */
2709                 nilfs_detach_log_writer(sb);
2710         }
2711
2712         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2713         if (!nilfs->ns_writer)
2714                 return -ENOMEM;
2715
2716         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2717         if (err) {
2718                 kfree(nilfs->ns_writer);
2719                 nilfs->ns_writer = NULL;
2720         }
2721         return err;
2722 }
2723
2724 /**
2725  * nilfs_detach_log_writer - destroy log writer
2726  * @sb: super block instance
2727  *
2728  * This kills log writer daemon, frees the log writer object, and
2729  * destroys list of dirty files.
2730  */
2731 void nilfs_detach_log_writer(struct super_block *sb)
2732 {
2733         struct the_nilfs *nilfs = sb->s_fs_info;
2734         LIST_HEAD(garbage_list);
2735
2736         down_write(&nilfs->ns_segctor_sem);
2737         if (nilfs->ns_writer) {
2738                 nilfs_segctor_destroy(nilfs->ns_writer);
2739                 nilfs->ns_writer = NULL;
2740         }
2741
2742         /* Force to free the list of dirty files */
2743         spin_lock(&nilfs->ns_inode_lock);
2744         if (!list_empty(&nilfs->ns_dirty_files)) {
2745                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2746                 nilfs_warning(sb, __func__,
2747                               "Hit dirty file after stopped log writer\n");
2748         }
2749         spin_unlock(&nilfs->ns_inode_lock);
2750         up_write(&nilfs->ns_segctor_sem);
2751
2752         nilfs_dispose_list(nilfs, &garbage_list, 1);
2753 }