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