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         INIT_LIST_HEAD(&ti->ti_garbage);
 306         current->journal_info = ti;
 307
 308         for (;;) {
 309                 down_write(&nilfs->ns_segctor_sem);
 310                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 311                         break;
 312
 313                 nilfs_segctor_do_immediate_flush(sci);
 314
 315                 up_write(&nilfs->ns_segctor_sem);
 316                 yield();
 317         }
 318         if (gcflag)
 319                 ti->ti_flags |= NILFS_TI_GC;
 320 }
 321
 322 static void nilfs_transaction_unlock(struct super_block *sb)
 323 {
 324         struct nilfs_transaction_info *ti = current->journal_info;
 325         struct the_nilfs *nilfs = sb->s_fs_info;
 326
 327         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 328         BUG_ON(ti->ti_count > 0);
 329
 330         up_write(&nilfs->ns_segctor_sem);
 331         current->journal_info = ti->ti_save;
 332         if (!list_empty(&ti->ti_garbage))
 333                 nilfs_dispose_list(nilfs, &ti->ti_garbage, 0);
 334 }
 335
 336 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 337                                             struct nilfs_segsum_pointer *ssp,
 338                                             unsigned bytes)
 339 {
 340         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 341         unsigned blocksize = sci->sc_super->s_blocksize;
 342         void *p;
 343
 344         if (unlikely(ssp->offset + bytes > blocksize)) {
 345                 ssp->offset = 0;
 346                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 347                                                &segbuf->sb_segsum_buffers));
 348                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 349         }
 350         p = ssp->bh->b_data + ssp->offset;
 351         ssp->offset += bytes;
 352         return p;
 353 }
 354
 355 /**
 356  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 357  * @sci: nilfs_sc_info
 358  */
 359 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 360 {
 361         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 362         struct buffer_head *sumbh;
 363         unsigned sumbytes;
 364         unsigned flags = 0;
 365         int err;
 366
 367         if (nilfs_doing_gc())
 368                 flags = NILFS_SS_GC;
 369         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 370         if (unlikely(err))
 371                 return err;
 372
 373         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 374         sumbytes = segbuf->sb_sum.sumbytes;
 375         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 376         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 377         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 378         return 0;
 379 }
 380
 381 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 382 {
 383         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 384         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 385                 return -E2BIG; /* The current segment is filled up
 386                                   (internal code) */
 387         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 388         return nilfs_segctor_reset_segment_buffer(sci);
 389 }
 390
 391 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 392 {
 393         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 394         int err;
 395
 396         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 397                 err = nilfs_segctor_feed_segment(sci);
 398                 if (err)
 399                         return err;
 400                 segbuf = sci->sc_curseg;
 401         }
 402         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 403         if (likely(!err))
 404                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 405         return err;
 406 }
 407
 408 /*
 409  * Functions for making segment summary and payloads
 410  */
 411 static int nilfs_segctor_segsum_block_required(
 412         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 413         unsigned binfo_size)
 414 {
 415         unsigned blocksize = sci->sc_super->s_blocksize;
 416         /* Size of finfo and binfo is enough small against blocksize */
 417
 418         return ssp->offset + binfo_size +
 419                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 420                 blocksize;
 421 }
 422
 423 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 424                                       struct inode *inode)
 425 {
 426         sci->sc_curseg->sb_sum.nfinfo++;
 427         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 428         nilfs_segctor_map_segsum_entry(
 429                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 430
 431         if (NILFS_I(inode)->i_root &&
 432             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 433                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 434         /* skip finfo */
 435 }
 436
 437 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 438                                     struct inode *inode)
 439 {
 440         struct nilfs_finfo *finfo;
 441         struct nilfs_inode_info *ii;
 442         struct nilfs_segment_buffer *segbuf;
 443         __u64 cno;
 444
 445         if (sci->sc_blk_cnt == 0)
 446                 return;
 447
 448         ii = NILFS_I(inode);
 449
 450         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 451                 cno = ii->i_cno;
 452         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 453                 cno = 0;
 454         else
 455                 cno = sci->sc_cno;
 456
 457         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 458                                                  sizeof(*finfo));
 459         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 460         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 461         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 462         finfo->fi_cno = cpu_to_le64(cno);
 463
 464         segbuf = sci->sc_curseg;
 465         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 466                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 467         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 468         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 469 }
 470
 471 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 472                                         struct buffer_head *bh,
 473                                         struct inode *inode,
 474                                         unsigned binfo_size)
 475 {
 476         struct nilfs_segment_buffer *segbuf;
 477         int required, err = 0;
 478
 479  retry:
 480         segbuf = sci->sc_curseg;
 481         required = nilfs_segctor_segsum_block_required(
 482                 sci, &sci->sc_binfo_ptr, binfo_size);
 483         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 484                 nilfs_segctor_end_finfo(sci, inode);
 485                 err = nilfs_segctor_feed_segment(sci);
 486                 if (err)
 487                         return err;
 488                 goto retry;
 489         }
 490         if (unlikely(required)) {
 491                 err = nilfs_segbuf_extend_segsum(segbuf);
 492                 if (unlikely(err))
 493                         goto failed;
 494         }
 495         if (sci->sc_blk_cnt == 0)
 496                 nilfs_segctor_begin_finfo(sci, inode);
 497
 498         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 499         /* Substitution to vblocknr is delayed until update_blocknr() */
 500         nilfs_segbuf_add_file_buffer(segbuf, bh);
 501         sci->sc_blk_cnt++;
 502  failed:
 503         return err;
 504 }
 505
 506 /*
 507  * Callback functions that enumerate, mark, and collect dirty blocks
 508  */
 509 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 510                                    struct buffer_head *bh, struct inode *inode)
 511 {
 512         int err;
 513
 514         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 515         if (err < 0)
 516                 return err;
 517
 518         err = nilfs_segctor_add_file_block(sci, bh, inode,
 519                                            sizeof(struct nilfs_binfo_v));
 520         if (!err)
 521                 sci->sc_datablk_cnt++;
 522         return err;
 523 }
 524
 525 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 526                                    struct buffer_head *bh,
 527                                    struct inode *inode)
 528 {
 529         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 530 }
 531
 532 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 533                                    struct buffer_head *bh,
 534                                    struct inode *inode)
 535 {
 536         WARN_ON(!buffer_dirty(bh));
 537         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 538 }
 539
 540 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 541                                         struct nilfs_segsum_pointer *ssp,
 542                                         union nilfs_binfo *binfo)
 543 {
 544         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 545                 sci, ssp, sizeof(*binfo_v));
 546         *binfo_v = binfo->bi_v;
 547 }
 548
 549 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 550                                         struct nilfs_segsum_pointer *ssp,
 551                                         union nilfs_binfo *binfo)
 552 {
 553         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 554                 sci, ssp, sizeof(*vblocknr));
 555         *vblocknr = binfo->bi_v.bi_vblocknr;
 556 }
 557
 558 static struct nilfs_sc_operations nilfs_sc_file_ops = {
 559         .collect_data = nilfs_collect_file_data,
 560         .collect_node = nilfs_collect_file_node,
 561         .collect_bmap = nilfs_collect_file_bmap,
 562         .write_data_binfo = nilfs_write_file_data_binfo,
 563         .write_node_binfo = nilfs_write_file_node_binfo,
 564 };
 565
 566 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 567                                   struct buffer_head *bh, struct inode *inode)
 568 {
 569         int err;
 570
 571         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 572         if (err < 0)
 573                 return err;
 574
 575         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 576         if (!err)
 577                 sci->sc_datablk_cnt++;
 578         return err;
 579 }
 580
 581 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 582                                   struct buffer_head *bh, struct inode *inode)
 583 {
 584         WARN_ON(!buffer_dirty(bh));
 585         return nilfs_segctor_add_file_block(sci, bh, inode,
 586                                             sizeof(struct nilfs_binfo_dat));
 587 }
 588
 589 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 590                                        struct nilfs_segsum_pointer *ssp,
 591                                        union nilfs_binfo *binfo)
 592 {
 593         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 594                                                           sizeof(*blkoff));
 595         *blkoff = binfo->bi_dat.bi_blkoff;
 596 }
 597
 598 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 599                                        struct nilfs_segsum_pointer *ssp,
 600                                        union nilfs_binfo *binfo)
 601 {
 602         struct nilfs_binfo_dat *binfo_dat =
 603                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 604         *binfo_dat = binfo->bi_dat;
 605 }
 606
 607 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
 608         .collect_data = nilfs_collect_dat_data,
 609         .collect_node = nilfs_collect_file_node,
 610         .collect_bmap = nilfs_collect_dat_bmap,
 611         .write_data_binfo = nilfs_write_dat_data_binfo,
 612         .write_node_binfo = nilfs_write_dat_node_binfo,
 613 };
 614
 615 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 616         .collect_data = nilfs_collect_file_data,
 617         .collect_node = NULL,
 618         .collect_bmap = NULL,
 619         .write_data_binfo = nilfs_write_file_data_binfo,
 620         .write_node_binfo = NULL,
 621 };
 622
 623 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 624                                               struct list_head *listp,
 625                                               size_t nlimit,
 626                                               loff_t start, loff_t end)
 627 {
 628         struct address_space *mapping = inode->i_mapping;
 629         struct pagevec pvec;
 630         pgoff_t index = 0, last = ULONG_MAX;
 631         size_t ndirties = 0;
 632         int i;
 633
 634         if (unlikely(start != 0 || end != LLONG_MAX)) {
 635                 /*
 636                  * A valid range is given for sync-ing data pages. The
 637                  * range is rounded to per-page; extra dirty buffers
 638                  * may be included if blocksize < pagesize.
 639                  */
 640                 index = start >> PAGE_SHIFT;
 641                 last = end >> PAGE_SHIFT;
 642         }
 643         pagevec_init(&pvec, 0);
 644  repeat:
 645         if (unlikely(index > last) ||
 646             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 647                                 min_t(pgoff_t, last - index,
 648                                       PAGEVEC_SIZE - 1) + 1))
 649                 return ndirties;
 650
 651         for (i = 0; i < pagevec_count(&pvec); i++) {
 652                 struct buffer_head *bh, *head;
 653                 struct page *page = pvec.pages[i];
 654
 655                 if (unlikely(page->index > last))
 656                         break;
 657
 658                 if (mapping->host) {
 659                         lock_page(page);
 660                         if (!page_has_buffers(page))
 661                                 create_empty_buffers(page,
 662                                                      1 << inode->i_blkbits, 0);
 663                         unlock_page(page);
 664                 }
 665
 666                 bh = head = page_buffers(page);
 667                 do {
 668                         if (!buffer_dirty(bh))
 669                                 continue;
 670                         get_bh(bh);
 671                         list_add_tail(&bh->b_assoc_buffers, listp);
 672                         ndirties++;
 673                         if (unlikely(ndirties >= nlimit)) {
 674                                 pagevec_release(&pvec);
 675                                 cond_resched();
 676                                 return ndirties;
 677                         }
 678                 } while (bh = bh->b_this_page, bh != head);
 679         }
 680         pagevec_release(&pvec);
 681         cond_resched();
 682         goto repeat;
 683 }
 684
 685 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 686                                             struct list_head *listp)
 687 {
 688         struct nilfs_inode_info *ii = NILFS_I(inode);
 689         struct address_space *mapping = &ii->i_btnode_cache;
 690         struct pagevec pvec;
 691         struct buffer_head *bh, *head;
 692         unsigned int i;
 693         pgoff_t index = 0;
 694
 695         pagevec_init(&pvec, 0);
 696
 697         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 698                                   PAGEVEC_SIZE)) {
 699                 for (i = 0; i < pagevec_count(&pvec); i++) {
 700                         bh = head = page_buffers(pvec.pages[i]);
 701                         do {
 702                                 if (buffer_dirty(bh)) {
 703                                         get_bh(bh);
 704                                         list_add_tail(&bh->b_assoc_buffers,
 705                                                       listp);
 706                                 }
 707                                 bh = bh->b_this_page;
 708                         } while (bh != head);
 709                 }
 710                 pagevec_release(&pvec);
 711                 cond_resched();
 712         }
 713 }
 714
 715 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 716                                struct list_head *head, int force)
 717 {
 718         struct nilfs_inode_info *ii, *n;
 719         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 720         unsigned nv = 0;
 721
 722         while (!list_empty(head)) {
 723                 spin_lock(&nilfs->ns_inode_lock);
 724                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 725                         list_del_init(&ii->i_dirty);
 726                         if (force) {
 727                                 if (unlikely(ii->i_bh)) {
 728                                         brelse(ii->i_bh);
 729                                         ii->i_bh = NULL;
 730                                 }
 731                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 732                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 733                                 list_add_tail(&ii->i_dirty,
 734                                               &nilfs->ns_dirty_files);
 735                                 continue;
 736                         }
 737                         ivec[nv++] = ii;
 738                         if (nv == SC_N_INODEVEC)
 739                                 break;
 740                 }
 741                 spin_unlock(&nilfs->ns_inode_lock);
 742
 743                 for (pii = ivec; nv > 0; pii++, nv--)
 744                         iput(&(*pii)->vfs_inode);
 745         }
 746 }
 747
 748 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 749                                      struct nilfs_root *root)
 750 {
 751         int ret = 0;
 752
 753         if (nilfs_mdt_fetch_dirty(root->ifile))
 754                 ret++;
 755         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 756                 ret++;
 757         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 758                 ret++;
 759         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 760                 ret++;
 761         return ret;
 762 }
 763
 764 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 765 {
 766         return list_empty(&sci->sc_dirty_files) &&
 767                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 768                 sci->sc_nfreesegs == 0 &&
 769                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 770 }
 771
 772 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 773 {
 774         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 775         int ret = 0;
 776
 777         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 778                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 779
 780         spin_lock(&nilfs->ns_inode_lock);
 781         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 782                 ret++;
 783
 784         spin_unlock(&nilfs->ns_inode_lock);
 785         return ret;
 786 }
 787
 788 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 789 {
 790         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 791
 792         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 793         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 794         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 795         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 796 }
 797
 798 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 799 {
 800         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 801         struct buffer_head *bh_cp;
 802         struct nilfs_checkpoint *raw_cp;
 803         int err;
 804
 805         /* XXX: this interface will be changed */
 806         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 807                                           &raw_cp, &bh_cp);
 808         if (likely(!err)) {
 809                 /* The following code is duplicated with cpfile.  But, it is
 810                    needed to collect the checkpoint even if it was not newly
 811                    created */
 812                 nilfs_mdt_mark_buffer_dirty(bh_cp);
 813                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 814                 nilfs_cpfile_put_checkpoint(
 815                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 816         } else
 817                 WARN_ON(err == -EINVAL || err == -ENOENT);
 818
 819         return err;
 820 }
 821
 822 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 823 {
 824         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 825         struct buffer_head *bh_cp;
 826         struct nilfs_checkpoint *raw_cp;
 827         int err;
 828
 829         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 830                                           &raw_cp, &bh_cp);
 831         if (unlikely(err)) {
 832                 WARN_ON(err == -EINVAL || err == -ENOENT);
 833                 goto failed_ibh;
 834         }
 835         raw_cp->cp_snapshot_list.ssl_next = 0;
 836         raw_cp->cp_snapshot_list.ssl_prev = 0;
 837         raw_cp->cp_inodes_count =
 838                 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
 839         raw_cp->cp_blocks_count =
 840                 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
 841         raw_cp->cp_nblk_inc =
 842                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 843         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 844         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 845
 846         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 847                 nilfs_checkpoint_clear_minor(raw_cp);
 848         else
 849                 nilfs_checkpoint_set_minor(raw_cp);
 850
 851         nilfs_write_inode_common(sci->sc_root->ifile,
 852                                  &raw_cp->cp_ifile_inode, 1);
 853         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 854         return 0;
 855
 856  failed_ibh:
 857         return err;
 858 }
 859
 860 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 861                                     struct nilfs_inode_info *ii)
 862
 863 {
 864         struct buffer_head *ibh;
 865         struct nilfs_inode *raw_inode;
 866
 867         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 868                 ibh = ii->i_bh;
 869                 BUG_ON(!ibh);
 870                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 871                                                   ibh);
 872                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 873                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 874         }
 875 }
 876
 877 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 878 {
 879         struct nilfs_inode_info *ii;
 880
 881         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 882                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 883                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 884         }
 885 }
 886
 887 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 888                                              struct the_nilfs *nilfs)
 889 {
 890         struct buffer_head *bh_sr;
 891         struct nilfs_super_root *raw_sr;
 892         unsigned isz = nilfs->ns_inode_size;
 893
 894         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 895         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 896
 897         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
 898         raw_sr->sr_nongc_ctime
 899                 = cpu_to_le64(nilfs_doing_gc() ?
 900                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 901         raw_sr->sr_flags = 0;
 902
 903         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 904                                  NILFS_SR_DAT_OFFSET(isz), 1);
 905         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 906                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 907         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 908                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 909 }
 910
 911 static void nilfs_redirty_inodes(struct list_head *head)
 912 {
 913         struct nilfs_inode_info *ii;
 914
 915         list_for_each_entry(ii, head, i_dirty) {
 916                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 917                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 918         }
 919 }
 920
 921 static void nilfs_drop_collected_inodes(struct list_head *head)
 922 {
 923         struct nilfs_inode_info *ii;
 924
 925         list_for_each_entry(ii, head, i_dirty) {
 926                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 927                         continue;
 928
 929                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 930                 set_bit(NILFS_I_UPDATED, &ii->i_state);
 931         }
 932 }
 933
 934 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
 935                                        struct inode *inode,
 936                                        struct list_head *listp,
 937                                        int (*collect)(struct nilfs_sc_info *,
 938                                                       struct buffer_head *,
 939                                                       struct inode *))
 940 {
 941         struct buffer_head *bh, *n;
 942         int err = 0;
 943
 944         if (collect) {
 945                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
 946                         list_del_init(&bh->b_assoc_buffers);
 947                         err = collect(sci, bh, inode);
 948                         brelse(bh);
 949                         if (unlikely(err))
 950                                 goto dispose_buffers;
 951                 }
 952                 return 0;
 953         }
 954
 955  dispose_buffers:
 956         while (!list_empty(listp)) {
 957                 bh = list_entry(listp->next, struct buffer_head,
 958                                 b_assoc_buffers);
 959                 list_del_init(&bh->b_assoc_buffers);
 960                 brelse(bh);
 961         }
 962         return err;
 963 }
 964
 965 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
 966 {
 967         /* Remaining number of blocks within segment buffer */
 968         return sci->sc_segbuf_nblocks -
 969                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
 970 }
 971
 972 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
 973                                    struct inode *inode,
 974                                    struct nilfs_sc_operations *sc_ops)
 975 {
 976         LIST_HEAD(data_buffers);
 977         LIST_HEAD(node_buffers);
 978         int err;
 979
 980         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
 981                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
 982
 983                 n = nilfs_lookup_dirty_data_buffers(
 984                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
 985                 if (n > rest) {
 986                         err = nilfs_segctor_apply_buffers(
 987                                 sci, inode, &data_buffers,
 988                                 sc_ops->collect_data);
 989                         BUG_ON(!err); /* always receive -E2BIG or true error */
 990                         goto break_or_fail;
 991                 }
 992         }
 993         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
 994
 995         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
 996                 err = nilfs_segctor_apply_buffers(
 997                         sci, inode, &data_buffers, sc_ops->collect_data);
 998                 if (unlikely(err)) {
 999                         /* dispose node list */
1000                         nilfs_segctor_apply_buffers(
1001                                 sci, inode, &node_buffers, NULL);
1002                         goto break_or_fail;
1003                 }
1004                 sci->sc_stage.flags |= NILFS_CF_NODE;
1005         }
1006         /* Collect node */
1007         err = nilfs_segctor_apply_buffers(
1008                 sci, inode, &node_buffers, sc_ops->collect_node);
1009         if (unlikely(err))
1010                 goto break_or_fail;
1011
1012         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1013         err = nilfs_segctor_apply_buffers(
1014                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1015         if (unlikely(err))
1016                 goto break_or_fail;
1017
1018         nilfs_segctor_end_finfo(sci, inode);
1019         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1020
1021  break_or_fail:
1022         return err;
1023 }
1024
1025 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1026                                          struct inode *inode)
1027 {
1028         LIST_HEAD(data_buffers);
1029         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1030         int err;
1031
1032         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1033                                             sci->sc_dsync_start,
1034                                             sci->sc_dsync_end);
1035
1036         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1037                                           nilfs_collect_file_data);
1038         if (!err) {
1039                 nilfs_segctor_end_finfo(sci, inode);
1040                 BUG_ON(n > rest);
1041                 /* always receive -E2BIG or true error if n > rest */
1042         }
1043         return err;
1044 }
1045
1046 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1047 {
1048         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1049         struct list_head *head;
1050         struct nilfs_inode_info *ii;
1051         size_t ndone;
1052         int err = 0;
1053
1054         switch (sci->sc_stage.scnt) {
1055         case NILFS_ST_INIT:
1056                 /* Pre-processes */
1057                 sci->sc_stage.flags = 0;
1058
1059                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1060                         sci->sc_nblk_inc = 0;
1061                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1062                         if (mode == SC_LSEG_DSYNC) {
1063                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1064                                 goto dsync_mode;
1065                         }
1066                 }
1067
1068                 sci->sc_stage.dirty_file_ptr = NULL;
1069                 sci->sc_stage.gc_inode_ptr = NULL;
1070                 if (mode == SC_FLUSH_DAT) {
1071                         sci->sc_stage.scnt = NILFS_ST_DAT;
1072                         goto dat_stage;
1073                 }
1074                 sci->sc_stage.scnt++;  /* Fall through */
1075         case NILFS_ST_GC:
1076                 if (nilfs_doing_gc()) {
1077                         head = &sci->sc_gc_inodes;
1078                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1079                                                 head, i_dirty);
1080                         list_for_each_entry_continue(ii, head, i_dirty) {
1081                                 err = nilfs_segctor_scan_file(
1082                                         sci, &ii->vfs_inode,
1083                                         &nilfs_sc_file_ops);
1084                                 if (unlikely(err)) {
1085                                         sci->sc_stage.gc_inode_ptr = list_entry(
1086                                                 ii->i_dirty.prev,
1087                                                 struct nilfs_inode_info,
1088                                                 i_dirty);
1089                                         goto break_or_fail;
1090                                 }
1091                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1092                         }
1093                         sci->sc_stage.gc_inode_ptr = NULL;
1094                 }
1095                 sci->sc_stage.scnt++;  /* Fall through */
1096         case NILFS_ST_FILE:
1097                 head = &sci->sc_dirty_files;
1098                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1099                                         i_dirty);
1100                 list_for_each_entry_continue(ii, head, i_dirty) {
1101                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1102
1103                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1104                                                       &nilfs_sc_file_ops);
1105                         if (unlikely(err)) {
1106                                 sci->sc_stage.dirty_file_ptr =
1107                                         list_entry(ii->i_dirty.prev,
1108                                                    struct nilfs_inode_info,
1109                                                    i_dirty);
1110                                 goto break_or_fail;
1111                         }
1112                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1113                         /* XXX: required ? */
1114                 }
1115                 sci->sc_stage.dirty_file_ptr = NULL;
1116                 if (mode == SC_FLUSH_FILE) {
1117                         sci->sc_stage.scnt = NILFS_ST_DONE;
1118                         return 0;
1119                 }
1120                 sci->sc_stage.scnt++;
1121                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1122                 /* Fall through */
1123         case NILFS_ST_IFILE:
1124                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1125                                               &nilfs_sc_file_ops);
1126                 if (unlikely(err))
1127                         break;
1128                 sci->sc_stage.scnt++;
1129                 /* Creating a checkpoint */
1130                 err = nilfs_segctor_create_checkpoint(sci);
1131                 if (unlikely(err))
1132                         break;
1133                 /* Fall through */
1134         case NILFS_ST_CPFILE:
1135                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1136                                               &nilfs_sc_file_ops);
1137                 if (unlikely(err))
1138                         break;
1139                 sci->sc_stage.scnt++;  /* Fall through */
1140         case NILFS_ST_SUFILE:
1141                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1142                                          sci->sc_nfreesegs, &ndone);
1143                 if (unlikely(err)) {
1144                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1145                                                   sci->sc_freesegs, ndone,
1146                                                   NULL);
1147                         break;
1148                 }
1149                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1150
1151                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1152                                               &nilfs_sc_file_ops);
1153                 if (unlikely(err))
1154                         break;
1155                 sci->sc_stage.scnt++;  /* Fall through */
1156         case NILFS_ST_DAT:
1157  dat_stage:
1158                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1159                                               &nilfs_sc_dat_ops);
1160                 if (unlikely(err))
1161                         break;
1162                 if (mode == SC_FLUSH_DAT) {
1163                         sci->sc_stage.scnt = NILFS_ST_DONE;
1164                         return 0;
1165                 }
1166                 sci->sc_stage.scnt++;  /* Fall through */
1167         case NILFS_ST_SR:
1168                 if (mode == SC_LSEG_SR) {
1169                         /* Appending a super root */
1170                         err = nilfs_segctor_add_super_root(sci);
1171                         if (unlikely(err))
1172                                 break;
1173                 }
1174                 /* End of a logical segment */
1175                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1176                 sci->sc_stage.scnt = NILFS_ST_DONE;
1177                 return 0;
1178         case NILFS_ST_DSYNC:
1179  dsync_mode:
1180                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1181                 ii = sci->sc_dsync_inode;
1182                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1183                         break;
1184
1185                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1186                 if (unlikely(err))
1187                         break;
1188                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1189                 sci->sc_stage.scnt = NILFS_ST_DONE;
1190                 return 0;
1191         case NILFS_ST_DONE:
1192                 return 0;
1193         default:
1194                 BUG();
1195         }
1196
1197  break_or_fail:
1198         return err;
1199 }
1200
1201 /**
1202  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1203  * @sci: nilfs_sc_info
1204  * @nilfs: nilfs object
1205  */
1206 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1207                                             struct the_nilfs *nilfs)
1208 {
1209         struct nilfs_segment_buffer *segbuf, *prev;
1210         __u64 nextnum;
1211         int err, alloc = 0;
1212
1213         segbuf = nilfs_segbuf_new(sci->sc_super);
1214         if (unlikely(!segbuf))
1215                 return -ENOMEM;
1216
1217         if (list_empty(&sci->sc_write_logs)) {
1218                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1219                                  nilfs->ns_pseg_offset, nilfs);
1220                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1221                         nilfs_shift_to_next_segment(nilfs);
1222                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1223                 }
1224
1225                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1226                 nextnum = nilfs->ns_nextnum;
1227
1228                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1229                         /* Start from the head of a new full segment */
1230                         alloc++;
1231         } else {
1232                 /* Continue logs */
1233                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1234                 nilfs_segbuf_map_cont(segbuf, prev);
1235                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1236                 nextnum = prev->sb_nextnum;
1237
1238                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1239                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1240                         segbuf->sb_sum.seg_seq++;
1241                         alloc++;
1242                 }
1243         }
1244
1245         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1246         if (err)
1247                 goto failed;
1248
1249         if (alloc) {
1250                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1251                 if (err)
1252                         goto failed;
1253         }
1254         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1255
1256         BUG_ON(!list_empty(&sci->sc_segbufs));
1257         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1258         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1259         return 0;
1260
1261  failed:
1262         nilfs_segbuf_free(segbuf);
1263         return err;
1264 }
1265
1266 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1267                                          struct the_nilfs *nilfs, int nadd)
1268 {
1269         struct nilfs_segment_buffer *segbuf, *prev;
1270         struct inode *sufile = nilfs->ns_sufile;
1271         __u64 nextnextnum;
1272         LIST_HEAD(list);
1273         int err, ret, i;
1274
1275         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1276         /*
1277          * Since the segment specified with nextnum might be allocated during
1278          * the previous construction, the buffer including its segusage may
1279          * not be dirty.  The following call ensures that the buffer is dirty
1280          * and will pin the buffer on memory until the sufile is written.
1281          */
1282         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1283         if (unlikely(err))
1284                 return err;
1285
1286         for (i = 0; i < nadd; i++) {
1287                 /* extend segment info */
1288                 err = -ENOMEM;
1289                 segbuf = nilfs_segbuf_new(sci->sc_super);
1290                 if (unlikely(!segbuf))
1291                         goto failed;
1292
1293                 /* map this buffer to region of segment on-disk */
1294                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1295                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1296
1297                 /* allocate the next next full segment */
1298                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1299                 if (unlikely(err))
1300                         goto failed_segbuf;
1301
1302                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1303                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1304
1305                 list_add_tail(&segbuf->sb_list, &list);
1306                 prev = segbuf;
1307         }
1308         list_splice_tail(&list, &sci->sc_segbufs);
1309         return 0;
1310
1311  failed_segbuf:
1312         nilfs_segbuf_free(segbuf);
1313  failed:
1314         list_for_each_entry(segbuf, &list, sb_list) {
1315                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1316                 WARN_ON(ret); /* never fails */
1317         }
1318         nilfs_destroy_logs(&list);
1319         return err;
1320 }
1321
1322 static void nilfs_free_incomplete_logs(struct list_head *logs,
1323                                        struct the_nilfs *nilfs)
1324 {
1325         struct nilfs_segment_buffer *segbuf, *prev;
1326         struct inode *sufile = nilfs->ns_sufile;
1327         int ret;
1328
1329         segbuf = NILFS_FIRST_SEGBUF(logs);
1330         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1331                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1332                 WARN_ON(ret); /* never fails */
1333         }
1334         if (atomic_read(&segbuf->sb_err)) {
1335                 /* Case 1: The first segment failed */
1336                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1337                         /* Case 1a:  Partial segment appended into an existing
1338                            segment */
1339                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1340                                                 segbuf->sb_fseg_end);
1341                 else /* Case 1b:  New full segment */
1342                         set_nilfs_discontinued(nilfs);
1343         }
1344
1345         prev = segbuf;
1346         list_for_each_entry_continue(segbuf, logs, sb_list) {
1347                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1348                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1349                         WARN_ON(ret); /* never fails */
1350                 }
1351                 if (atomic_read(&segbuf->sb_err) &&
1352                     segbuf->sb_segnum != nilfs->ns_nextnum)
1353                         /* Case 2: extended segment (!= next) failed */
1354                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1355                 prev = segbuf;
1356         }
1357 }
1358
1359 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1360                                           struct inode *sufile)
1361 {
1362         struct nilfs_segment_buffer *segbuf;
1363         unsigned long live_blocks;
1364         int ret;
1365
1366         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1367                 live_blocks = segbuf->sb_sum.nblocks +
1368                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1369                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1370                                                      live_blocks,
1371                                                      sci->sc_seg_ctime);
1372                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1373         }
1374 }
1375
1376 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1377 {
1378         struct nilfs_segment_buffer *segbuf;
1379         int ret;
1380
1381         segbuf = NILFS_FIRST_SEGBUF(logs);
1382         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1383                                              segbuf->sb_pseg_start -
1384                                              segbuf->sb_fseg_start, 0);
1385         WARN_ON(ret); /* always succeed because the segusage is dirty */
1386
1387         list_for_each_entry_continue(segbuf, logs, sb_list) {
1388                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1389                                                      0, 0);
1390                 WARN_ON(ret); /* always succeed */
1391         }
1392 }
1393
1394 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1395                                             struct nilfs_segment_buffer *last,
1396                                             struct inode *sufile)
1397 {
1398         struct nilfs_segment_buffer *segbuf = last;
1399         int ret;
1400
1401         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1402                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1403                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1404                 WARN_ON(ret);
1405         }
1406         nilfs_truncate_logs(&sci->sc_segbufs, last);
1407 }
1408
1409
1410 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1411                                  struct the_nilfs *nilfs, int mode)
1412 {
1413         struct nilfs_cstage prev_stage = sci->sc_stage;
1414         int err, nadd = 1;
1415
1416         /* Collection retry loop */
1417         for (;;) {
1418                 sci->sc_nblk_this_inc = 0;
1419                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1420
1421                 err = nilfs_segctor_reset_segment_buffer(sci);
1422                 if (unlikely(err))
1423                         goto failed;
1424
1425                 err = nilfs_segctor_collect_blocks(sci, mode);
1426                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1427                 if (!err)
1428                         break;
1429
1430                 if (unlikely(err != -E2BIG))
1431                         goto failed;
1432
1433                 /* The current segment is filled up */
1434                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1435                         break;
1436
1437                 nilfs_clear_logs(&sci->sc_segbufs);
1438
1439                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1440                 if (unlikely(err))
1441                         return err;
1442
1443                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1444                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1445                                                         sci->sc_freesegs,
1446                                                         sci->sc_nfreesegs,
1447                                                         NULL);
1448                         WARN_ON(err); /* do not happen */
1449                 }
1450                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1451                 sci->sc_stage = prev_stage;
1452         }
1453         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1454         return 0;
1455
1456  failed:
1457         return err;
1458 }
1459
1460 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1461                                       struct buffer_head *new_bh)
1462 {
1463         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1464
1465         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1466         /* The caller must release old_bh */
1467 }
1468
1469 static int
1470 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1471                                      struct nilfs_segment_buffer *segbuf,
1472                                      int mode)
1473 {
1474         struct inode *inode = NULL;
1475         sector_t blocknr;
1476         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1477         unsigned long nblocks = 0, ndatablk = 0;
1478         struct nilfs_sc_operations *sc_op = NULL;
1479         struct nilfs_segsum_pointer ssp;
1480         struct nilfs_finfo *finfo = NULL;
1481         union nilfs_binfo binfo;
1482         struct buffer_head *bh, *bh_org;
1483         ino_t ino = 0;
1484         int err = 0;
1485
1486         if (!nfinfo)
1487                 goto out;
1488
1489         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1490         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1491         ssp.offset = sizeof(struct nilfs_segment_summary);
1492
1493         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1494                 if (bh == segbuf->sb_super_root)
1495                         break;
1496                 if (!finfo) {
1497                         finfo = nilfs_segctor_map_segsum_entry(
1498                                 sci, &ssp, sizeof(*finfo));
1499                         ino = le64_to_cpu(finfo->fi_ino);
1500                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1501                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1502
1503                         if (buffer_nilfs_node(bh))
1504                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1505                         else
1506                                 inode = NILFS_AS_I(bh->b_page->mapping);
1507
1508                         if (mode == SC_LSEG_DSYNC)
1509                                 sc_op = &nilfs_sc_dsync_ops;
1510                         else if (ino == NILFS_DAT_INO)
1511                                 sc_op = &nilfs_sc_dat_ops;
1512                         else /* file blocks */
1513                                 sc_op = &nilfs_sc_file_ops;
1514                 }
1515                 bh_org = bh;
1516                 get_bh(bh_org);
1517                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1518                                         &binfo);
1519                 if (bh != bh_org)
1520                         nilfs_list_replace_buffer(bh_org, bh);
1521                 brelse(bh_org);
1522                 if (unlikely(err))
1523                         goto failed_bmap;
1524
1525                 if (ndatablk > 0)
1526                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1527                 else
1528                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1529
1530                 blocknr++;
1531                 if (--nblocks == 0) {
1532                         finfo = NULL;
1533                         if (--nfinfo == 0)
1534                                 break;
1535                 } else if (ndatablk > 0)
1536                         ndatablk--;
1537         }
1538  out:
1539         return 0;
1540
1541  failed_bmap:
1542         return err;
1543 }
1544
1545 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1546 {
1547         struct nilfs_segment_buffer *segbuf;
1548         int err;
1549
1550         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1551                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1552                 if (unlikely(err))
1553                         return err;
1554                 nilfs_segbuf_fill_in_segsum(segbuf);
1555         }
1556         return 0;
1557 }
1558
1559 static int
1560 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1561 {
1562         struct page *clone_page;
1563         struct buffer_head *bh, *head, *bh2;
1564         void *kaddr;
1565
1566         bh = head = page_buffers(page);
1567
1568         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1569         if (unlikely(!clone_page))
1570                 return -ENOMEM;
1571
1572         bh2 = page_buffers(clone_page);
1573         kaddr = kmap_atomic(page, KM_USER0);
1574         do {
1575                 if (list_empty(&bh->b_assoc_buffers))
1576                         continue;
1577                 get_bh(bh2);
1578                 page_cache_get(clone_page); /* for each bh */
1579                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1580                 bh2->b_blocknr = bh->b_blocknr;
1581                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1582                 list_add_tail(&bh->b_assoc_buffers, out);
1583         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1584         kunmap_atomic(kaddr, KM_USER0);
1585
1586         if (!TestSetPageWriteback(clone_page))
1587                 account_page_writeback(clone_page);
1588         unlock_page(clone_page);
1589
1590         return 0;
1591 }
1592
1593 static int nilfs_test_page_to_be_frozen(struct page *page)
1594 {
1595         struct address_space *mapping = page->mapping;
1596
1597         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1598                 return 0;
1599
1600         if (page_mapped(page)) {
1601                 ClearPageChecked(page);
1602                 return 1;
1603         }
1604         return PageChecked(page);
1605 }
1606
1607 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1608 {
1609         if (!page || PageWriteback(page))
1610                 /* For split b-tree node pages, this function may be called
1611                    twice.  We ignore the 2nd or later calls by this check. */
1612                 return 0;
1613
1614         lock_page(page);
1615         clear_page_dirty_for_io(page);
1616         set_page_writeback(page);
1617         unlock_page(page);
1618
1619         if (nilfs_test_page_to_be_frozen(page)) {
1620                 int err = nilfs_copy_replace_page_buffers(page, out);
1621                 if (unlikely(err))
1622                         return err;
1623         }
1624         return 0;
1625 }
1626
1627 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1628                                        struct page **failed_page)
1629 {
1630         struct nilfs_segment_buffer *segbuf;
1631         struct page *bd_page = NULL, *fs_page = NULL;
1632         struct list_head *list = &sci->sc_copied_buffers;
1633         int err;
1634
1635         *failed_page = NULL;
1636         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1637                 struct buffer_head *bh;
1638
1639                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1640                                     b_assoc_buffers) {
1641                         if (bh->b_page != bd_page) {
1642                                 if (bd_page) {
1643                                         lock_page(bd_page);
1644                                         clear_page_dirty_for_io(bd_page);
1645                                         set_page_writeback(bd_page);
1646                                         unlock_page(bd_page);
1647                                 }
1648                                 bd_page = bh->b_page;
1649                         }
1650                 }
1651
1652                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1653                                     b_assoc_buffers) {
1654                         if (bh == segbuf->sb_super_root) {
1655                                 if (bh->b_page != bd_page) {
1656                                         lock_page(bd_page);
1657                                         clear_page_dirty_for_io(bd_page);
1658                                         set_page_writeback(bd_page);
1659                                         unlock_page(bd_page);
1660                                         bd_page = bh->b_page;
1661                                 }
1662                                 break;
1663                         }
1664                         if (bh->b_page != fs_page) {
1665                                 err = nilfs_begin_page_io(fs_page, list);
1666                                 if (unlikely(err)) {
1667                                         *failed_page = fs_page;
1668                                         goto out;
1669                                 }
1670                                 fs_page = bh->b_page;
1671                         }
1672                 }
1673         }
1674         if (bd_page) {
1675                 lock_page(bd_page);
1676                 clear_page_dirty_for_io(bd_page);
1677                 set_page_writeback(bd_page);
1678                 unlock_page(bd_page);
1679         }
1680         err = nilfs_begin_page_io(fs_page, list);
1681         if (unlikely(err))
1682                 *failed_page = fs_page;
1683  out:
1684         return err;
1685 }
1686
1687 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1688                                struct the_nilfs *nilfs)
1689 {
1690         int ret;
1691
1692         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1693         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1694         return ret;
1695 }
1696
1697 static void __nilfs_end_page_io(struct page *page, int err)
1698 {
1699         if (!err) {
1700                 if (!nilfs_page_buffers_clean(page))
1701                         __set_page_dirty_nobuffers(page);
1702                 ClearPageError(page);
1703         } else {
1704                 __set_page_dirty_nobuffers(page);
1705                 SetPageError(page);
1706         }
1707
1708         if (buffer_nilfs_allocated(page_buffers(page))) {
1709                 if (TestClearPageWriteback(page))
1710                         dec_zone_page_state(page, NR_WRITEBACK);
1711         } else
1712                 end_page_writeback(page);
1713 }
1714
1715 static void nilfs_end_page_io(struct page *page, int err)
1716 {
1717         if (!page)
1718                 return;
1719
1720         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1721                 /*
1722                  * For b-tree node pages, this function may be called twice
1723                  * or more because they might be split in a segment.
1724                  */
1725                 if (PageDirty(page)) {
1726                         /*
1727                          * For pages holding split b-tree node buffers, dirty
1728                          * flag on the buffers may be cleared discretely.
1729                          * In that case, the page is once redirtied for
1730                          * remaining buffers, and it must be cancelled if
1731                          * all the buffers get cleaned later.
1732                          */
1733                         lock_page(page);
1734                         if (nilfs_page_buffers_clean(page))
1735                                 __nilfs_clear_page_dirty(page);
1736                         unlock_page(page);
1737                 }
1738                 return;
1739         }
1740
1741         __nilfs_end_page_io(page, err);
1742 }
1743
1744 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1745 {
1746         struct buffer_head *bh, *head;
1747         struct page *page;
1748
1749         while (!list_empty(list)) {
1750                 bh = list_entry(list->next, struct buffer_head,
1751                                 b_assoc_buffers);
1752                 page = bh->b_page;
1753                 page_cache_get(page);
1754                 head = bh = page_buffers(page);
1755                 do {
1756                         if (!list_empty(&bh->b_assoc_buffers)) {
1757                                 list_del_init(&bh->b_assoc_buffers);
1758                                 if (!err) {
1759                                         set_buffer_uptodate(bh);
1760                                         clear_buffer_dirty(bh);
1761                                         clear_buffer_delay(bh);
1762                                         clear_buffer_nilfs_volatile(bh);
1763                                 }
1764                                 brelse(bh); /* for b_assoc_buffers */
1765                         }
1766                 } while ((bh = bh->b_this_page) != head);
1767
1768                 __nilfs_end_page_io(page, err);
1769                 page_cache_release(page);
1770         }
1771 }
1772
1773 static void nilfs_abort_logs(struct list_head *logs, struct page *failed_page,
1774                              int err)
1775 {
1776         struct nilfs_segment_buffer *segbuf;
1777         struct page *bd_page = NULL, *fs_page = NULL;
1778         struct buffer_head *bh;
1779
1780         if (list_empty(logs))
1781                 return;
1782
1783         list_for_each_entry(segbuf, logs, sb_list) {
1784                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1785                                     b_assoc_buffers) {
1786                         if (bh->b_page != bd_page) {
1787                                 if (bd_page)
1788                                         end_page_writeback(bd_page);
1789                                 bd_page = bh->b_page;
1790                         }
1791                 }
1792
1793                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1794                                     b_assoc_buffers) {
1795                         if (bh == segbuf->sb_super_root) {
1796                                 if (bh->b_page != bd_page) {
1797                                         end_page_writeback(bd_page);
1798                                         bd_page = bh->b_page;
1799                                 }
1800                                 break;
1801                         }
1802                         if (bh->b_page != fs_page) {
1803                                 nilfs_end_page_io(fs_page, err);
1804                                 if (fs_page && fs_page == failed_page)
1805                                         return;
1806                                 fs_page = bh->b_page;
1807                         }
1808                 }
1809         }
1810         if (bd_page)
1811                 end_page_writeback(bd_page);
1812
1813         nilfs_end_page_io(fs_page, err);
1814 }
1815
1816 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1817                                              struct the_nilfs *nilfs, int err)
1818 {
1819         LIST_HEAD(logs);
1820         int ret;
1821
1822         list_splice_tail_init(&sci->sc_write_logs, &logs);
1823         ret = nilfs_wait_on_logs(&logs);
1824         nilfs_abort_logs(&logs, NULL, ret ? : err);
1825
1826         list_splice_tail_init(&sci->sc_segbufs, &logs);
1827         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1828         nilfs_free_incomplete_logs(&logs, nilfs);
1829         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1830
1831         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1832                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1833                                                 sci->sc_freesegs,
1834                                                 sci->sc_nfreesegs,
1835                                                 NULL);
1836                 WARN_ON(ret); /* do not happen */
1837         }
1838
1839         nilfs_destroy_logs(&logs);
1840 }
1841
1842 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1843                                    struct nilfs_segment_buffer *segbuf)
1844 {
1845         nilfs->ns_segnum = segbuf->sb_segnum;
1846         nilfs->ns_nextnum = segbuf->sb_nextnum;
1847         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1848                 + segbuf->sb_sum.nblocks;
1849         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1850         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1851 }
1852
1853 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1854 {
1855         struct nilfs_segment_buffer *segbuf;
1856         struct page *bd_page = NULL, *fs_page = NULL;
1857         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1858         int update_sr = false;
1859
1860         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1861                 struct buffer_head *bh;
1862
1863                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1864                                     b_assoc_buffers) {
1865                         set_buffer_uptodate(bh);
1866                         clear_buffer_dirty(bh);
1867                         if (bh->b_page != bd_page) {
1868                                 if (bd_page)
1869                                         end_page_writeback(bd_page);
1870                                 bd_page = bh->b_page;
1871                         }
1872                 }
1873                 /*
1874                  * We assume that the buffers which belong to the same page
1875                  * continue over the buffer list.
1876                  * Under this assumption, the last BHs of pages is
1877                  * identifiable by the discontinuity of bh->b_page
1878                  * (page != fs_page).
1879                  *
1880                  * For B-tree node blocks, however, this assumption is not
1881                  * guaranteed.  The cleanup code of B-tree node pages needs
1882                  * special care.
1883                  */
1884                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1885                                     b_assoc_buffers) {
1886                         set_buffer_uptodate(bh);
1887                         clear_buffer_dirty(bh);
1888                         clear_buffer_delay(bh);
1889                         clear_buffer_nilfs_volatile(bh);
1890                         clear_buffer_nilfs_redirected(bh);
1891                         if (bh == segbuf->sb_super_root) {
1892                                 if (bh->b_page != bd_page) {
1893                                         end_page_writeback(bd_page);
1894                                         bd_page = bh->b_page;
1895                                 }
1896                                 update_sr = true;
1897                                 break;
1898                         }
1899                         if (bh->b_page != fs_page) {
1900                                 nilfs_end_page_io(fs_page, 0);
1901                                 fs_page = bh->b_page;
1902                         }
1903                 }
1904
1905                 if (!nilfs_segbuf_simplex(segbuf)) {
1906                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1907                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1908                                 sci->sc_lseg_stime = jiffies;
1909                         }
1910                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1911                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1912                 }
1913         }
1914         /*
1915          * Since pages may continue over multiple segment buffers,
1916          * end of the last page must be checked outside of the loop.
1917          */
1918         if (bd_page)
1919                 end_page_writeback(bd_page);
1920
1921         nilfs_end_page_io(fs_page, 0);
1922
1923         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1924
1925         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1926
1927         if (nilfs_doing_gc())
1928                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1929         else
1930                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1931
1932         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1933
1934         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1935         nilfs_set_next_segment(nilfs, segbuf);
1936
1937         if (update_sr) {
1938                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1939                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1940
1941                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1942                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1943                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1944                 nilfs_segctor_clear_metadata_dirty(sci);
1945         } else
1946                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1947 }
1948
1949 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1950 {
1951         int ret;
1952
1953         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1954         if (!ret) {
1955                 nilfs_segctor_complete_write(sci);
1956                 nilfs_destroy_logs(&sci->sc_write_logs);
1957         }
1958         return ret;
1959 }
1960
1961 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1962                                              struct the_nilfs *nilfs)
1963 {
1964         struct nilfs_inode_info *ii, *n;
1965         struct inode *ifile = sci->sc_root->ifile;
1966
1967         spin_lock(&nilfs->ns_inode_lock);
1968  retry:
1969         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1970                 if (!ii->i_bh) {
1971                         struct buffer_head *ibh;
1972                         int err;
1973
1974                         spin_unlock(&nilfs->ns_inode_lock);
1975                         err = nilfs_ifile_get_inode_block(
1976                                 ifile, ii->vfs_inode.i_ino, &ibh);
1977                         if (unlikely(err)) {
1978                                 nilfs_warning(sci->sc_super, __func__,
1979                                               "failed to get inode block.\n");
1980                                 return err;
1981                         }
1982                         nilfs_mdt_mark_buffer_dirty(ibh);
1983                         nilfs_mdt_mark_dirty(ifile);
1984                         spin_lock(&nilfs->ns_inode_lock);
1985                         if (likely(!ii->i_bh))
1986                                 ii->i_bh = ibh;
1987                         else
1988                                 brelse(ibh);
1989                         goto retry;
1990                 }
1991
1992                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1993                 set_bit(NILFS_I_BUSY, &ii->i_state);
1994                 list_del(&ii->i_dirty);
1995                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
1996         }
1997         spin_unlock(&nilfs->ns_inode_lock);
1998
1999         return 0;
2000 }
2001
2002 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
2003                                              struct the_nilfs *nilfs)
2004 {
2005         struct nilfs_transaction_info *ti = current->journal_info;
2006         struct nilfs_inode_info *ii, *n;
2007
2008         spin_lock(&nilfs->ns_inode_lock);
2009         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2010                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2011                     test_bit(NILFS_I_DIRTY, &ii->i_state))
2012                         continue;
2013
2014                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2015                 brelse(ii->i_bh);
2016                 ii->i_bh = NULL;
2017                 list_del(&ii->i_dirty);
2018                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2019         }
2020         spin_unlock(&nilfs->ns_inode_lock);
2021 }
2022
2023 /*
2024  * Main procedure of segment constructor
2025  */
2026 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2027 {
2028         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2029         struct page *failed_page;
2030         int err;
2031
2032         sci->sc_stage.scnt = NILFS_ST_INIT;
2033         sci->sc_cno = nilfs->ns_cno;
2034
2035         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2036         if (unlikely(err))
2037                 goto out;
2038
2039         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2040                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2041
2042         if (nilfs_segctor_clean(sci))
2043                 goto out;
2044
2045         do {
2046                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2047
2048                 err = nilfs_segctor_begin_construction(sci, nilfs);
2049                 if (unlikely(err))
2050                         goto out;
2051
2052                 /* Update time stamp */
2053                 sci->sc_seg_ctime = get_seconds();
2054
2055                 err = nilfs_segctor_collect(sci, nilfs, mode);
2056                 if (unlikely(err))
2057                         goto failed;
2058
2059                 /* Avoid empty segment */
2060                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2061                     nilfs_segbuf_empty(sci->sc_curseg)) {
2062                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2063                         goto out;
2064                 }
2065
2066                 err = nilfs_segctor_assign(sci, mode);
2067                 if (unlikely(err))
2068                         goto failed;
2069
2070                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2071                         nilfs_segctor_fill_in_file_bmap(sci);
2072
2073                 if (mode == SC_LSEG_SR &&
2074                     sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
2075                         err = nilfs_segctor_fill_in_checkpoint(sci);
2076                         if (unlikely(err))
2077                                 goto failed_to_write;
2078
2079                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2080                 }
2081                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2082
2083                 /* Write partial segments */
2084                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2085                 if (err) {
2086                         nilfs_abort_logs(&sci->sc_segbufs, failed_page, err);
2087                         goto failed_to_write;
2088                 }
2089
2090                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2091                                             nilfs->ns_crc_seed);
2092
2093                 err = nilfs_segctor_write(sci, nilfs);
2094                 if (unlikely(err))
2095                         goto failed_to_write;
2096
2097                 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2098                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2099                         /*
2100                          * At this point, we avoid double buffering
2101                          * for blocksize < pagesize because page dirty
2102                          * flag is turned off during write and dirty
2103                          * buffers are not properly collected for
2104                          * pages crossing over segments.
2105                          */
2106                         err = nilfs_segctor_wait(sci);
2107                         if (err)
2108                                 goto failed_to_write;
2109                 }
2110         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2111
2112  out:
2113         nilfs_segctor_drop_written_files(sci, nilfs);
2114         return err;
2115
2116  failed_to_write:
2117         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2118                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2119
2120  failed:
2121         if (nilfs_doing_gc())
2122                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2123         nilfs_segctor_abort_construction(sci, nilfs, err);
2124         goto out;
2125 }
2126
2127 /**
2128  * nilfs_segctor_start_timer - set timer of background write
2129  * @sci: nilfs_sc_info
2130  *
2131  * If the timer has already been set, it ignores the new request.
2132  * This function MUST be called within a section locking the segment
2133  * semaphore.
2134  */
2135 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2136 {
2137         spin_lock(&sci->sc_state_lock);
2138         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2139                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2140                 add_timer(&sci->sc_timer);
2141                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2142         }
2143         spin_unlock(&sci->sc_state_lock);
2144 }
2145
2146 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2147 {
2148         spin_lock(&sci->sc_state_lock);
2149         if (!(sci->sc_flush_request & (1 << bn))) {
2150                 unsigned long prev_req = sci->sc_flush_request;
2151
2152                 sci->sc_flush_request |= (1 << bn);
2153                 if (!prev_req)
2154                         wake_up(&sci->sc_wait_daemon);
2155         }
2156         spin_unlock(&sci->sc_state_lock);
2157 }
2158
2159 /**
2160  * nilfs_flush_segment - trigger a segment construction for resource control
2161  * @sb: super block
2162  * @ino: inode number of the file to be flushed out.
2163  */
2164 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2165 {
2166         struct the_nilfs *nilfs = sb->s_fs_info;
2167         struct nilfs_sc_info *sci = nilfs->ns_writer;
2168
2169         if (!sci || nilfs_doing_construction())
2170                 return;
2171         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2172                                         /* assign bit 0 to data files */
2173 }
2174
2175 struct nilfs_segctor_wait_request {
2176         wait_queue_t    wq;
2177         __u32           seq;
2178         int             err;
2179         atomic_t        done;
2180 };
2181
2182 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2183 {
2184         struct nilfs_segctor_wait_request wait_req;
2185         int err = 0;
2186
2187         spin_lock(&sci->sc_state_lock);
2188         init_wait(&wait_req.wq);
2189         wait_req.err = 0;
2190         atomic_set(&wait_req.done, 0);
2191         wait_req.seq = ++sci->sc_seq_request;
2192         spin_unlock(&sci->sc_state_lock);
2193
2194         init_waitqueue_entry(&wait_req.wq, current);
2195         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2196         set_current_state(TASK_INTERRUPTIBLE);
2197         wake_up(&sci->sc_wait_daemon);
2198
2199         for (;;) {
2200                 if (atomic_read(&wait_req.done)) {
2201                         err = wait_req.err;
2202                         break;
2203                 }
2204                 if (!signal_pending(current)) {
2205                         schedule();
2206                         continue;
2207                 }
2208                 err = -ERESTARTSYS;
2209                 break;
2210         }
2211         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2212         return err;
2213 }
2214
2215 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2216 {
2217         struct nilfs_segctor_wait_request *wrq, *n;
2218         unsigned long flags;
2219
2220         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2221         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2222                                  wq.task_list) {
2223                 if (!atomic_read(&wrq->done) &&
2224                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2225                         wrq->err = err;
2226                         atomic_set(&wrq->done, 1);
2227                 }
2228                 if (atomic_read(&wrq->done)) {
2229                         wrq->wq.func(&wrq->wq,
2230                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2231                                      0, NULL);
2232                 }
2233         }
2234         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2235 }
2236
2237 /**
2238  * nilfs_construct_segment - construct a logical segment
2239  * @sb: super block
2240  *
2241  * Return Value: On success, 0 is retured. On errors, one of the following
2242  * negative error code is returned.
2243  *
2244  * %-EROFS - Read only filesystem.
2245  *
2246  * %-EIO - I/O error
2247  *
2248  * %-ENOSPC - No space left on device (only in a panic state).
2249  *
2250  * %-ERESTARTSYS - Interrupted.
2251  *
2252  * %-ENOMEM - Insufficient memory available.
2253  */
2254 int nilfs_construct_segment(struct super_block *sb)
2255 {
2256         struct the_nilfs *nilfs = sb->s_fs_info;
2257         struct nilfs_sc_info *sci = nilfs->ns_writer;
2258         struct nilfs_transaction_info *ti;
2259         int err;
2260
2261         if (!sci)
2262                 return -EROFS;
2263
2264         /* A call inside transactions causes a deadlock. */
2265         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2266
2267         err = nilfs_segctor_sync(sci);
2268         return err;
2269 }
2270
2271 /**
2272  * nilfs_construct_dsync_segment - construct a data-only logical segment
2273  * @sb: super block
2274  * @inode: inode whose data blocks should be written out
2275  * @start: start byte offset
2276  * @end: end byte offset (inclusive)
2277  *
2278  * Return Value: On success, 0 is retured. On errors, one of the following
2279  * negative error code is returned.
2280  *
2281  * %-EROFS - Read only filesystem.
2282  *
2283  * %-EIO - I/O error
2284  *
2285  * %-ENOSPC - No space left on device (only in a panic state).
2286  *
2287  * %-ERESTARTSYS - Interrupted.
2288  *
2289  * %-ENOMEM - Insufficient memory available.
2290  */
2291 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2292                                   loff_t start, loff_t end)
2293 {
2294         struct the_nilfs *nilfs = sb->s_fs_info;
2295         struct nilfs_sc_info *sci = nilfs->ns_writer;
2296         struct nilfs_inode_info *ii;
2297         struct nilfs_transaction_info ti;
2298         int err = 0;
2299
2300         if (!sci)
2301                 return -EROFS;
2302
2303         nilfs_transaction_lock(sb, &ti, 0);
2304
2305         ii = NILFS_I(inode);
2306         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2307             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2308             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2309             nilfs_discontinued(nilfs)) {
2310                 nilfs_transaction_unlock(sb);
2311                 err = nilfs_segctor_sync(sci);
2312                 return err;
2313         }
2314
2315         spin_lock(&nilfs->ns_inode_lock);
2316         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2317             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2318                 spin_unlock(&nilfs->ns_inode_lock);
2319                 nilfs_transaction_unlock(sb);
2320                 return 0;
2321         }
2322         spin_unlock(&nilfs->ns_inode_lock);
2323         sci->sc_dsync_inode = ii;
2324         sci->sc_dsync_start = start;
2325         sci->sc_dsync_end = end;
2326
2327         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2328
2329         nilfs_transaction_unlock(sb);
2330         return err;
2331 }
2332
2333 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2334 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2335
2336 /**
2337  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2338  * @sci: segment constructor object
2339  */
2340 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2341 {
2342         spin_lock(&sci->sc_state_lock);
2343         sci->sc_seq_accepted = sci->sc_seq_request;
2344         spin_unlock(&sci->sc_state_lock);
2345         del_timer_sync(&sci->sc_timer);
2346 }
2347
2348 /**
2349  * nilfs_segctor_notify - notify the result of request to caller threads
2350  * @sci: segment constructor object
2351  * @mode: mode of log forming
2352  * @err: error code to be notified
2353  */
2354 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2355 {
2356         /* Clear requests (even when the construction failed) */
2357         spin_lock(&sci->sc_state_lock);
2358
2359         if (mode == SC_LSEG_SR) {
2360                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2361                 sci->sc_seq_done = sci->sc_seq_accepted;
2362                 nilfs_segctor_wakeup(sci, err);
2363                 sci->sc_flush_request = 0;
2364         } else {
2365                 if (mode == SC_FLUSH_FILE)
2366                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2367                 else if (mode == SC_FLUSH_DAT)
2368                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2369
2370                 /* re-enable timer if checkpoint creation was not done */
2371                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2372                     time_before(jiffies, sci->sc_timer.expires))
2373                         add_timer(&sci->sc_timer);
2374         }
2375         spin_unlock(&sci->sc_state_lock);
2376 }
2377
2378 /**
2379  * nilfs_segctor_construct - form logs and write them to disk
2380  * @sci: segment constructor object
2381  * @mode: mode of log forming
2382  */
2383 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2384 {
2385         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2386         struct nilfs_super_block **sbp;
2387         int err = 0;
2388
2389         nilfs_segctor_accept(sci);
2390
2391         if (nilfs_discontinued(nilfs))
2392                 mode = SC_LSEG_SR;
2393         if (!nilfs_segctor_confirm(sci))
2394                 err = nilfs_segctor_do_construct(sci, mode);
2395
2396         if (likely(!err)) {
2397                 if (mode != SC_FLUSH_DAT)
2398                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2399                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2400                     nilfs_discontinued(nilfs)) {
2401                         down_write(&nilfs->ns_sem);
2402                         err = -EIO;
2403                         sbp = nilfs_prepare_super(sci->sc_super,
2404                                                   nilfs_sb_will_flip(nilfs));
2405                         if (likely(sbp)) {
2406                                 nilfs_set_log_cursor(sbp[0], nilfs);
2407                                 err = nilfs_commit_super(sci->sc_super,
2408                                                          NILFS_SB_COMMIT);
2409                         }
2410                         up_write(&nilfs->ns_sem);
2411                 }
2412         }
2413
2414         nilfs_segctor_notify(sci, mode, err);
2415         return err;
2416 }
2417
2418 static void nilfs_construction_timeout(unsigned long data)
2419 {
2420         struct task_struct *p = (struct task_struct *)data;
2421         wake_up_process(p);
2422 }
2423
2424 static void
2425 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2426 {
2427         struct nilfs_inode_info *ii, *n;
2428
2429         list_for_each_entry_safe(ii, n, head, i_dirty) {
2430                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2431                         continue;
2432                 list_del_init(&ii->i_dirty);
2433                 iput(&ii->vfs_inode);
2434         }
2435 }
2436
2437 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2438                          void **kbufs)
2439 {
2440         struct the_nilfs *nilfs = sb->s_fs_info;
2441         struct nilfs_sc_info *sci = nilfs->ns_writer;
2442         struct nilfs_transaction_info ti;
2443         int err;
2444
2445         if (unlikely(!sci))
2446                 return -EROFS;
2447
2448         nilfs_transaction_lock(sb, &ti, 1);
2449
2450         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2451         if (unlikely(err))
2452                 goto out_unlock;
2453
2454         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2455         if (unlikely(err)) {
2456                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2457                 goto out_unlock;
2458         }
2459
2460         sci->sc_freesegs = kbufs[4];
2461         sci->sc_nfreesegs = argv[4].v_nmembs;
2462         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2463
2464         for (;;) {
2465                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2466                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2467
2468                 if (likely(!err))
2469                         break;
2470
2471                 nilfs_warning(sb, __func__,
2472                               "segment construction failed. (err=%d)", err);
2473                 set_current_state(TASK_INTERRUPTIBLE);
2474                 schedule_timeout(sci->sc_interval);
2475         }
2476         if (nilfs_test_opt(nilfs, DISCARD)) {
2477                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2478                                                  sci->sc_nfreesegs);
2479                 if (ret) {
2480                         printk(KERN_WARNING
2481                                "NILFS warning: error %d on discard request, "
2482                                "turning discards off for the device\n", ret);
2483                         nilfs_clear_opt(nilfs, DISCARD);
2484                 }
2485         }
2486
2487  out_unlock:
2488         sci->sc_freesegs = NULL;
2489         sci->sc_nfreesegs = 0;
2490         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2491         nilfs_transaction_unlock(sb);
2492         return err;
2493 }
2494
2495 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2496 {
2497         struct nilfs_transaction_info ti;
2498
2499         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2500         nilfs_segctor_construct(sci, mode);
2501
2502         /*
2503          * Unclosed segment should be retried.  We do this using sc_timer.
2504          * Timeout of sc_timer will invoke complete construction which leads
2505          * to close the current logical segment.
2506          */
2507         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2508                 nilfs_segctor_start_timer(sci);
2509
2510         nilfs_transaction_unlock(sci->sc_super);
2511 }
2512
2513 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2514 {
2515         int mode = 0;
2516         int err;
2517
2518         spin_lock(&sci->sc_state_lock);
2519         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2520                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2521         spin_unlock(&sci->sc_state_lock);
2522
2523         if (mode) {
2524                 err = nilfs_segctor_do_construct(sci, mode);
2525
2526                 spin_lock(&sci->sc_state_lock);
2527                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2528                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2529                 spin_unlock(&sci->sc_state_lock);
2530         }
2531         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2532 }
2533
2534 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2535 {
2536         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2537             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2538                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2539                         return SC_FLUSH_FILE;
2540                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2541                         return SC_FLUSH_DAT;
2542         }
2543         return SC_LSEG_SR;
2544 }
2545
2546 /**
2547  * nilfs_segctor_thread - main loop of the segment constructor thread.
2548  * @arg: pointer to a struct nilfs_sc_info.
2549  *
2550  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2551  * to execute segment constructions.
2552  */
2553 static int nilfs_segctor_thread(void *arg)
2554 {
2555         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2556         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2557         int timeout = 0;
2558
2559         sci->sc_timer.data = (unsigned long)current;
2560         sci->sc_timer.function = nilfs_construction_timeout;
2561
2562         /* start sync. */
2563         sci->sc_task = current;
2564         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2565         printk(KERN_INFO
2566                "segctord starting. Construction interval = %lu seconds, "
2567                "CP frequency < %lu seconds\n",
2568                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2569
2570         spin_lock(&sci->sc_state_lock);
2571  loop:
2572         for (;;) {
2573                 int mode;
2574
2575                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2576                         goto end_thread;
2577
2578                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2579                         mode = SC_LSEG_SR;
2580                 else if (!sci->sc_flush_request)
2581                         break;
2582                 else
2583                         mode = nilfs_segctor_flush_mode(sci);
2584
2585                 spin_unlock(&sci->sc_state_lock);
2586                 nilfs_segctor_thread_construct(sci, mode);
2587                 spin_lock(&sci->sc_state_lock);
2588                 timeout = 0;
2589         }
2590
2591
2592         if (freezing(current)) {
2593                 spin_unlock(&sci->sc_state_lock);
2594                 refrigerator();
2595                 spin_lock(&sci->sc_state_lock);
2596         } else {
2597                 DEFINE_WAIT(wait);
2598                 int should_sleep = 1;
2599
2600                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2601                                 TASK_INTERRUPTIBLE);
2602
2603                 if (sci->sc_seq_request != sci->sc_seq_done)
2604                         should_sleep = 0;
2605                 else if (sci->sc_flush_request)
2606                         should_sleep = 0;
2607                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2608                         should_sleep = time_before(jiffies,
2609                                         sci->sc_timer.expires);
2610
2611                 if (should_sleep) {
2612                         spin_unlock(&sci->sc_state_lock);
2613                         schedule();
2614                         spin_lock(&sci->sc_state_lock);
2615                 }
2616                 finish_wait(&sci->sc_wait_daemon, &wait);
2617                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2618                            time_after_eq(jiffies, sci->sc_timer.expires));
2619
2620                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2621                         set_nilfs_discontinued(nilfs);
2622         }
2623         goto loop;
2624
2625  end_thread:
2626         spin_unlock(&sci->sc_state_lock);
2627
2628         /* end sync. */
2629         sci->sc_task = NULL;
2630         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2631         return 0;
2632 }
2633
2634 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2635 {
2636         struct task_struct *t;
2637
2638         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2639         if (IS_ERR(t)) {
2640                 int err = PTR_ERR(t);
2641
2642                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2643                        err);
2644                 return err;
2645         }
2646         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2647         return 0;
2648 }
2649
2650 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2651         __acquires(&sci->sc_state_lock)
2652         __releases(&sci->sc_state_lock)
2653 {
2654         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2655
2656         while (sci->sc_task) {
2657                 wake_up(&sci->sc_wait_daemon);
2658                 spin_unlock(&sci->sc_state_lock);
2659                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2660                 spin_lock(&sci->sc_state_lock);
2661         }
2662 }
2663
2664 /*
2665  * Setup & clean-up functions
2666  */
2667 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2668                                                struct nilfs_root *root)
2669 {
2670         struct the_nilfs *nilfs = sb->s_fs_info;
2671         struct nilfs_sc_info *sci;
2672
2673         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2674         if (!sci)
2675                 return NULL;
2676
2677         sci->sc_super = sb;
2678
2679         nilfs_get_root(root);
2680         sci->sc_root = root;
2681
2682         init_waitqueue_head(&sci->sc_wait_request);
2683         init_waitqueue_head(&sci->sc_wait_daemon);
2684         init_waitqueue_head(&sci->sc_wait_task);
2685         spin_lock_init(&sci->sc_state_lock);
2686         INIT_LIST_HEAD(&sci->sc_dirty_files);
2687         INIT_LIST_HEAD(&sci->sc_segbufs);
2688         INIT_LIST_HEAD(&sci->sc_write_logs);
2689         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2690         INIT_LIST_HEAD(&sci->sc_copied_buffers);
2691         init_timer(&sci->sc_timer);
2692
2693         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2694         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2695         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2696
2697         if (nilfs->ns_interval)
2698                 sci->sc_interval = nilfs->ns_interval;
2699         if (nilfs->ns_watermark)
2700                 sci->sc_watermark = nilfs->ns_watermark;
2701         return sci;
2702 }
2703
2704 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2705 {
2706         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2707
2708         /* The segctord thread was stopped and its timer was removed.
2709            But some tasks remain. */
2710         do {
2711                 struct nilfs_transaction_info ti;
2712
2713                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2714                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2715                 nilfs_transaction_unlock(sci->sc_super);
2716
2717         } while (ret && retrycount-- > 0);
2718 }
2719
2720 /**
2721  * nilfs_segctor_destroy - destroy the segment constructor.
2722  * @sci: nilfs_sc_info
2723  *
2724  * nilfs_segctor_destroy() kills the segctord thread and frees
2725  * the nilfs_sc_info struct.
2726  * Caller must hold the segment semaphore.
2727  */
2728 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2729 {
2730         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2731         int flag;
2732
2733         up_write(&nilfs->ns_segctor_sem);
2734
2735         spin_lock(&sci->sc_state_lock);
2736         nilfs_segctor_kill_thread(sci);
2737         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2738                 || sci->sc_seq_request != sci->sc_seq_done);
2739         spin_unlock(&sci->sc_state_lock);
2740
2741         if (flag || !nilfs_segctor_confirm(sci))
2742                 nilfs_segctor_write_out(sci);
2743
2744         WARN_ON(!list_empty(&sci->sc_copied_buffers));
2745
2746         if (!list_empty(&sci->sc_dirty_files)) {
2747                 nilfs_warning(sci->sc_super, __func__,
2748                               "dirty file(s) after the final construction\n");
2749                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2750         }
2751
2752         WARN_ON(!list_empty(&sci->sc_segbufs));
2753         WARN_ON(!list_empty(&sci->sc_write_logs));
2754
2755         nilfs_put_root(sci->sc_root);
2756
2757         down_write(&nilfs->ns_segctor_sem);
2758
2759         del_timer_sync(&sci->sc_timer);
2760         kfree(sci);
2761 }
2762
2763 /**
2764  * nilfs_attach_log_writer - attach log writer
2765  * @sb: super block instance
2766  * @root: root object of the current filesystem tree
2767  *
2768  * This allocates a log writer object, initializes it, and starts the
2769  * log writer.
2770  *
2771  * Return Value: On success, 0 is returned. On error, one of the following
2772  * negative error code is returned.
2773  *
2774  * %-ENOMEM - Insufficient memory available.
2775  */
2776 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2777 {
2778         struct the_nilfs *nilfs = sb->s_fs_info;
2779         int err;
2780
2781         if (nilfs->ns_writer) {
2782                 /*
2783                  * This happens if the filesystem was remounted
2784                  * read/write after nilfs_error degenerated it into a
2785                  * read-only mount.
2786                  */
2787                 nilfs_detach_log_writer(sb);
2788         }
2789
2790         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2791         if (!nilfs->ns_writer)
2792                 return -ENOMEM;
2793
2794         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2795         if (err) {
2796                 kfree(nilfs->ns_writer);
2797                 nilfs->ns_writer = NULL;
2798         }
2799         return err;
2800 }
2801
2802 /**
2803  * nilfs_detach_log_writer - destroy log writer
2804  * @sb: super block instance
2805  *
2806  * This kills log writer daemon, frees the log writer object, and
2807  * destroys list of dirty files.
2808  */
2809 void nilfs_detach_log_writer(struct super_block *sb)
2810 {
2811         struct the_nilfs *nilfs = sb->s_fs_info;
2812         LIST_HEAD(garbage_list);
2813
2814         down_write(&nilfs->ns_segctor_sem);
2815         if (nilfs->ns_writer) {
2816                 nilfs_segctor_destroy(nilfs->ns_writer);
2817                 nilfs->ns_writer = NULL;
2818         }
2819
2820         /* Force to free the list of dirty files */
2821         spin_lock(&nilfs->ns_inode_lock);
2822         if (!list_empty(&nilfs->ns_dirty_files)) {
2823                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2824                 nilfs_warning(sb, __func__,
2825                               "Hit dirty file after stopped log writer\n");
2826         }
2827         spin_unlock(&nilfs->ns_inode_lock);
2828         up_write(&nilfs->ns_segctor_sem);
2829
2830         nilfs_dispose_list(nilfs, &garbage_list, 1);
2831 }