fs/nilfs2/inode.c

   1 /*
   2  * inode.c - NILFS inode operations.
   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/buffer_head.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mpage.h>
  27 #include <linux/writeback.h>
  28 #include <linux/uio.h>
  29 #include "nilfs.h"
  30 #include "btnode.h"
  31 #include "segment.h"
  32 #include "page.h"
  33 #include "mdt.h"
  34 #include "cpfile.h"
  35 #include "ifile.h"
  36
  37 struct nilfs_iget_args {
  38         u64 ino;
  39         __u64 cno;
  40         struct nilfs_root *root;
  41         int for_gc;
  42 };
  43
  44 void nilfs_inode_add_blocks(struct inode *inode, int n)
  45 {
  46         struct nilfs_root *root = NILFS_I(inode)->i_root;
  47
  48         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  49         if (root)
  50                 atomic_add(n, &root->blocks_count);
  51 }
  52
  53 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  54 {
  55         struct nilfs_root *root = NILFS_I(inode)->i_root;
  56
  57         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  58         if (root)
  59                 atomic_sub(n, &root->blocks_count);
  60 }
  61
  62 /**
  63  * nilfs_get_block() - get a file block on the filesystem (callback function)
  64  * @inode - inode struct of the target file
  65  * @blkoff - file block number
  66  * @bh_result - buffer head to be mapped on
  67  * @create - indicate whether allocating the block or not when it has not
  68  *      been allocated yet.
  69  *
  70  * This function does not issue actual read request of the specified data
  71  * block. It is done by VFS.
  72  */
  73 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  74                     struct buffer_head *bh_result, int create)
  75 {
  76         struct nilfs_inode_info *ii = NILFS_I(inode);
  77         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  78         __u64 blknum = 0;
  79         int err = 0, ret;
  80         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  81
  82         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  83         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  84         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  85         if (ret >= 0) { /* found */
  86                 map_bh(bh_result, inode->i_sb, blknum);
  87                 if (ret > 0)
  88                         bh_result->b_size = (ret << inode->i_blkbits);
  89                 goto out;
  90         }
  91         /* data block was not found */
  92         if (ret == -ENOENT && create) {
  93                 struct nilfs_transaction_info ti;
  94
  95                 bh_result->b_blocknr = 0;
  96                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
  97                 if (unlikely(err))
  98                         goto out;
  99                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 100                                         (unsigned long)bh_result);
 101                 if (unlikely(err != 0)) {
 102                         if (err == -EEXIST) {
 103                                 /*
 104                                  * The get_block() function could be called
 105                                  * from multiple callers for an inode.
 106                                  * However, the page having this block must
 107                                  * be locked in this case.
 108                                  */
 109                                 printk(KERN_WARNING
 110                                        "nilfs_get_block: a race condition "
 111                                        "while inserting a data block. "
 112                                        "(inode number=%lu, file block "
 113                                        "offset=%llu)\n",
 114                                        inode->i_ino,
 115                                        (unsigned long long)blkoff);
 116                                 err = 0;
 117                         }
 118                         nilfs_transaction_abort(inode->i_sb);
 119                         goto out;
 120                 }
 121                 nilfs_mark_inode_dirty(inode);
 122                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 123                 /* Error handling should be detailed */
 124                 set_buffer_new(bh_result);
 125                 set_buffer_delay(bh_result);
 126                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 127                                                       to proper value */
 128         } else if (ret == -ENOENT) {
 129                 /* not found is not error (e.g. hole); must return without
 130                    the mapped state flag. */
 131                 ;
 132         } else {
 133                 err = ret;
 134         }
 135
 136  out:
 137         return err;
 138 }
 139
 140 /**
 141  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 142  * address_space_operations.
 143  * @file - file struct of the file to be read
 144  * @page - the page to be read
 145  */
 146 static int nilfs_readpage(struct file *file, struct page *page)
 147 {
 148         return mpage_readpage(page, nilfs_get_block);
 149 }
 150
 151 /**
 152  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 153  * address_space_operations.
 154  * @file - file struct of the file to be read
 155  * @mapping - address_space struct used for reading multiple pages
 156  * @pages - the pages to be read
 157  * @nr_pages - number of pages to be read
 158  */
 159 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 160                            struct list_head *pages, unsigned nr_pages)
 161 {
 162         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 163 }
 164
 165 static int nilfs_writepages(struct address_space *mapping,
 166                             struct writeback_control *wbc)
 167 {
 168         struct inode *inode = mapping->host;
 169         int err = 0;
 170
 171         if (wbc->sync_mode == WB_SYNC_ALL)
 172                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 173                                                     wbc->range_start,
 174                                                     wbc->range_end);
 175         return err;
 176 }
 177
 178 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 179 {
 180         struct inode *inode = page->mapping->host;
 181         int err;
 182
 183         redirty_page_for_writepage(wbc, page);
 184         unlock_page(page);
 185
 186         if (wbc->sync_mode == WB_SYNC_ALL) {
 187                 err = nilfs_construct_segment(inode->i_sb);
 188                 if (unlikely(err))
 189                         return err;
 190         } else if (wbc->for_reclaim)
 191                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 192
 193         return 0;
 194 }
 195
 196 static int nilfs_set_page_dirty(struct page *page)
 197 {
 198         int ret = __set_page_dirty_buffers(page);
 199
 200         if (ret) {
 201                 struct inode *inode = page->mapping->host;
 202                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 203
 204                 nilfs_set_file_dirty(inode, nr_dirty);
 205         }
 206         return ret;
 207 }
 208
 209 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 210                              loff_t pos, unsigned len, unsigned flags,
 211                              struct page **pagep, void **fsdata)
 212
 213 {
 214         struct inode *inode = mapping->host;
 215         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 216
 217         if (unlikely(err))
 218                 return err;
 219
 220         err = block_write_begin(mapping, pos, len, flags, pagep,
 221                                 nilfs_get_block);
 222         if (unlikely(err)) {
 223                 loff_t isize = mapping->host->i_size;
 224                 if (pos + len > isize)
 225                         vmtruncate(mapping->host, isize);
 226
 227                 nilfs_transaction_abort(inode->i_sb);
 228         }
 229         return err;
 230 }
 231
 232 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 233                            loff_t pos, unsigned len, unsigned copied,
 234                            struct page *page, void *fsdata)
 235 {
 236         struct inode *inode = mapping->host;
 237         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 238         unsigned nr_dirty;
 239         int err;
 240
 241         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 242                                                   start + copied);
 243         copied = generic_write_end(file, mapping, pos, len, copied, page,
 244                                    fsdata);
 245         nilfs_set_file_dirty(inode, nr_dirty);
 246         err = nilfs_transaction_commit(inode->i_sb);
 247         return err ? : copied;
 248 }
 249
 250 static ssize_t
 251 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 252                 loff_t offset, unsigned long nr_segs)
 253 {
 254         struct file *file = iocb->ki_filp;
 255         struct inode *inode = file->f_mapping->host;
 256         ssize_t size;
 257
 258         if (rw == WRITE)
 259                 return 0;
 260
 261         /* Needs synchronization with the cleaner */
 262         size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 263                                   nilfs_get_block);
 264
 265         /*
 266          * In case of error extending write may have instantiated a few
 267          * blocks outside i_size. Trim these off again.
 268          */
 269         if (unlikely((rw & WRITE) && size < 0)) {
 270                 loff_t isize = i_size_read(inode);
 271                 loff_t end = offset + iov_length(iov, nr_segs);
 272
 273                 if (end > isize)
 274                         vmtruncate(inode, isize);
 275         }
 276
 277         return size;
 278 }
 279
 280 const struct address_space_operations nilfs_aops = {
 281         .writepage              = nilfs_writepage,
 282         .readpage               = nilfs_readpage,
 283         .writepages             = nilfs_writepages,
 284         .set_page_dirty         = nilfs_set_page_dirty,
 285         .readpages              = nilfs_readpages,
 286         .write_begin            = nilfs_write_begin,
 287         .write_end              = nilfs_write_end,
 288         /* .releasepage         = nilfs_releasepage, */
 289         .invalidatepage         = block_invalidatepage,
 290         .direct_IO              = nilfs_direct_IO,
 291         .is_partially_uptodate  = block_is_partially_uptodate,
 292 };
 293
 294 struct inode *nilfs_new_inode(struct inode *dir, int mode)
 295 {
 296         struct super_block *sb = dir->i_sb;
 297         struct the_nilfs *nilfs = sb->s_fs_info;
 298         struct inode *inode;
 299         struct nilfs_inode_info *ii;
 300         struct nilfs_root *root;
 301         int err = -ENOMEM;
 302         ino_t ino;
 303
 304         inode = new_inode(sb);
 305         if (unlikely(!inode))
 306                 goto failed;
 307
 308         mapping_set_gfp_mask(inode->i_mapping,
 309                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 310
 311         root = NILFS_I(dir)->i_root;
 312         ii = NILFS_I(inode);
 313         ii->i_state = 1 << NILFS_I_NEW;
 314         ii->i_root = root;
 315
 316         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 317         if (unlikely(err))
 318                 goto failed_ifile_create_inode;
 319         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 320
 321         atomic_inc(&root->inodes_count);
 322         inode_init_owner(inode, dir, mode);
 323         inode->i_ino = ino;
 324         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 325
 326         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 327                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 328                 if (err < 0)
 329                         goto failed_bmap;
 330
 331                 set_bit(NILFS_I_BMAP, &ii->i_state);
 332                 /* No lock is needed; iget() ensures it. */
 333         }
 334
 335         ii->i_flags = nilfs_mask_flags(
 336                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 337
 338         /* ii->i_file_acl = 0; */
 339         /* ii->i_dir_acl = 0; */
 340         ii->i_dir_start_lookup = 0;
 341         nilfs_set_inode_flags(inode);
 342         spin_lock(&nilfs->ns_next_gen_lock);
 343         inode->i_generation = nilfs->ns_next_generation++;
 344         spin_unlock(&nilfs->ns_next_gen_lock);
 345         insert_inode_hash(inode);
 346
 347         err = nilfs_init_acl(inode, dir);
 348         if (unlikely(err))
 349                 goto failed_acl; /* never occur. When supporting
 350                                     nilfs_init_acl(), proper cancellation of
 351                                     above jobs should be considered */
 352
 353         return inode;
 354
 355  failed_acl:
 356  failed_bmap:
 357         clear_nlink(inode);
 358         iput(inode);  /* raw_inode will be deleted through
 359                          generic_delete_inode() */
 360         goto failed;
 361
 362  failed_ifile_create_inode:
 363         make_bad_inode(inode);
 364         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 365                          called */
 366  failed:
 367         return ERR_PTR(err);
 368 }
 369
 370 void nilfs_set_inode_flags(struct inode *inode)
 371 {
 372         unsigned int flags = NILFS_I(inode)->i_flags;
 373
 374         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 375                             S_DIRSYNC);
 376         if (flags & FS_SYNC_FL)
 377                 inode->i_flags |= S_SYNC;
 378         if (flags & FS_APPEND_FL)
 379                 inode->i_flags |= S_APPEND;
 380         if (flags & FS_IMMUTABLE_FL)
 381                 inode->i_flags |= S_IMMUTABLE;
 382         if (flags & FS_NOATIME_FL)
 383                 inode->i_flags |= S_NOATIME;
 384         if (flags & FS_DIRSYNC_FL)
 385                 inode->i_flags |= S_DIRSYNC;
 386         mapping_set_gfp_mask(inode->i_mapping,
 387                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 388 }
 389
 390 int nilfs_read_inode_common(struct inode *inode,
 391                             struct nilfs_inode *raw_inode)
 392 {
 393         struct nilfs_inode_info *ii = NILFS_I(inode);
 394         int err;
 395
 396         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 397         inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
 398         inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
 399         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 400         inode->i_size = le64_to_cpu(raw_inode->i_size);
 401         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 402         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 403         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 404         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 405         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 406         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 407         if (inode->i_nlink == 0 && inode->i_mode == 0)
 408                 return -EINVAL; /* this inode is deleted */
 409
 410         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 411         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 412 #if 0
 413         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 414         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 415                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 416 #endif
 417         ii->i_dir_start_lookup = 0;
 418         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 419
 420         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 421             S_ISLNK(inode->i_mode)) {
 422                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 423                 if (err < 0)
 424                         return err;
 425                 set_bit(NILFS_I_BMAP, &ii->i_state);
 426                 /* No lock is needed; iget() ensures it. */
 427         }
 428         return 0;
 429 }
 430
 431 static int __nilfs_read_inode(struct super_block *sb,
 432                               struct nilfs_root *root, unsigned long ino,
 433                               struct inode *inode)
 434 {
 435         struct the_nilfs *nilfs = sb->s_fs_info;
 436         struct buffer_head *bh;
 437         struct nilfs_inode *raw_inode;
 438         int err;
 439
 440         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 441         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 442         if (unlikely(err))
 443                 goto bad_inode;
 444
 445         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 446
 447         err = nilfs_read_inode_common(inode, raw_inode);
 448         if (err)
 449                 goto failed_unmap;
 450
 451         if (S_ISREG(inode->i_mode)) {
 452                 inode->i_op = &nilfs_file_inode_operations;
 453                 inode->i_fop = &nilfs_file_operations;
 454                 inode->i_mapping->a_ops = &nilfs_aops;
 455         } else if (S_ISDIR(inode->i_mode)) {
 456                 inode->i_op = &nilfs_dir_inode_operations;
 457                 inode->i_fop = &nilfs_dir_operations;
 458                 inode->i_mapping->a_ops = &nilfs_aops;
 459         } else if (S_ISLNK(inode->i_mode)) {
 460                 inode->i_op = &nilfs_symlink_inode_operations;
 461                 inode->i_mapping->a_ops = &nilfs_aops;
 462         } else {
 463                 inode->i_op = &nilfs_special_inode_operations;
 464                 init_special_inode(
 465                         inode, inode->i_mode,
 466                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 467         }
 468         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 469         brelse(bh);
 470         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 471         nilfs_set_inode_flags(inode);
 472         return 0;
 473
 474  failed_unmap:
 475         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 476         brelse(bh);
 477
 478  bad_inode:
 479         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 480         return err;
 481 }
 482
 483 static int nilfs_iget_test(struct inode *inode, void *opaque)
 484 {
 485         struct nilfs_iget_args *args = opaque;
 486         struct nilfs_inode_info *ii;
 487
 488         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 489                 return 0;
 490
 491         ii = NILFS_I(inode);
 492         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 493                 return !args->for_gc;
 494
 495         return args->for_gc && args->cno == ii->i_cno;
 496 }
 497
 498 static int nilfs_iget_set(struct inode *inode, void *opaque)
 499 {
 500         struct nilfs_iget_args *args = opaque;
 501
 502         inode->i_ino = args->ino;
 503         if (args->for_gc) {
 504                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 505                 NILFS_I(inode)->i_cno = args->cno;
 506                 NILFS_I(inode)->i_root = NULL;
 507         } else {
 508                 if (args->root && args->ino == NILFS_ROOT_INO)
 509                         nilfs_get_root(args->root);
 510                 NILFS_I(inode)->i_root = args->root;
 511         }
 512         return 0;
 513 }
 514
 515 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 516                             unsigned long ino)
 517 {
 518         struct nilfs_iget_args args = {
 519                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 520         };
 521
 522         return ilookup5(sb, ino, nilfs_iget_test, &args);
 523 }
 524
 525 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 526                                 unsigned long ino)
 527 {
 528         struct nilfs_iget_args args = {
 529                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 530         };
 531
 532         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 533 }
 534
 535 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 536                          unsigned long ino)
 537 {
 538         struct inode *inode;
 539         int err;
 540
 541         inode = nilfs_iget_locked(sb, root, ino);
 542         if (unlikely(!inode))
 543                 return ERR_PTR(-ENOMEM);
 544         if (!(inode->i_state & I_NEW))
 545                 return inode;
 546
 547         err = __nilfs_read_inode(sb, root, ino, inode);
 548         if (unlikely(err)) {
 549                 iget_failed(inode);
 550                 return ERR_PTR(err);
 551         }
 552         unlock_new_inode(inode);
 553         return inode;
 554 }
 555
 556 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 557                                 __u64 cno)
 558 {
 559         struct nilfs_iget_args args = {
 560                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 561         };
 562         struct inode *inode;
 563         int err;
 564
 565         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 566         if (unlikely(!inode))
 567                 return ERR_PTR(-ENOMEM);
 568         if (!(inode->i_state & I_NEW))
 569                 return inode;
 570
 571         err = nilfs_init_gcinode(inode);
 572         if (unlikely(err)) {
 573                 iget_failed(inode);
 574                 return ERR_PTR(err);
 575         }
 576         unlock_new_inode(inode);
 577         return inode;
 578 }
 579
 580 void nilfs_write_inode_common(struct inode *inode,
 581                               struct nilfs_inode *raw_inode, int has_bmap)
 582 {
 583         struct nilfs_inode_info *ii = NILFS_I(inode);
 584
 585         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 586         raw_inode->i_uid = cpu_to_le32(inode->i_uid);
 587         raw_inode->i_gid = cpu_to_le32(inode->i_gid);
 588         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 589         raw_inode->i_size = cpu_to_le64(inode->i_size);
 590         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 591         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 592         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 593         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 594         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 595
 596         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 597         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 598
 599         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 600                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 601
 602                 /* zero-fill unused portion in the case of super root block */
 603                 raw_inode->i_xattr = 0;
 604                 raw_inode->i_pad = 0;
 605                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 606                        nilfs->ns_inode_size - sizeof(*raw_inode));
 607         }
 608
 609         if (has_bmap)
 610                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 611         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 612                 raw_inode->i_device_code =
 613                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 614         /* When extending inode, nilfs->ns_inode_size should be checked
 615            for substitutions of appended fields */
 616 }
 617
 618 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 619 {
 620         ino_t ino = inode->i_ino;
 621         struct nilfs_inode_info *ii = NILFS_I(inode);
 622         struct inode *ifile = ii->i_root->ifile;
 623         struct nilfs_inode *raw_inode;
 624
 625         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 626
 627         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 628                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 629         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 630
 631         nilfs_write_inode_common(inode, raw_inode, 0);
 632                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 633                    deadlock of bmap. This delays update of i_bmap to just
 634                    before writing */
 635         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 636 }
 637
 638 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 639
 640 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 641                                 unsigned long from)
 642 {
 643         unsigned long b;
 644         int ret;
 645
 646         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 647                 return;
 648 repeat:
 649         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 650         if (ret == -ENOENT)
 651                 return;
 652         else if (ret < 0)
 653                 goto failed;
 654
 655         if (b < from)
 656                 return;
 657
 658         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 659         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 660         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 661         if (!ret || (ret == -ENOMEM &&
 662                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 663                 goto repeat;
 664
 665 failed:
 666         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 667                       "failed to truncate bmap (ino=%lu, err=%d)",
 668                       ii->vfs_inode.i_ino, ret);
 669 }
 670
 671 void nilfs_truncate(struct inode *inode)
 672 {
 673         unsigned long blkoff;
 674         unsigned int blocksize;
 675         struct nilfs_transaction_info ti;
 676         struct super_block *sb = inode->i_sb;
 677         struct nilfs_inode_info *ii = NILFS_I(inode);
 678
 679         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 680                 return;
 681         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 682                 return;
 683
 684         blocksize = sb->s_blocksize;
 685         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 686         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 687
 688         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 689
 690         nilfs_truncate_bmap(ii, blkoff);
 691
 692         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 693         if (IS_SYNC(inode))
 694                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 695
 696         nilfs_mark_inode_dirty(inode);
 697         nilfs_set_file_dirty(inode, 0);
 698         nilfs_transaction_commit(sb);
 699         /* May construct a logical segment and may fail in sync mode.
 700            But truncate has no return value. */
 701 }
 702
 703 static void nilfs_clear_inode(struct inode *inode)
 704 {
 705         struct nilfs_inode_info *ii = NILFS_I(inode);
 706         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 707
 708         /*
 709          * Free resources allocated in nilfs_read_inode(), here.
 710          */
 711         BUG_ON(!list_empty(&ii->i_dirty));
 712         brelse(ii->i_bh);
 713         ii->i_bh = NULL;
 714
 715         if (mdi && mdi->mi_palloc_cache)
 716                 nilfs_palloc_destroy_cache(inode);
 717
 718         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 719                 nilfs_bmap_clear(ii->i_bmap);
 720
 721         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 722
 723         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 724                 nilfs_put_root(ii->i_root);
 725 }
 726
 727 void nilfs_evict_inode(struct inode *inode)
 728 {
 729         struct nilfs_transaction_info ti;
 730         struct super_block *sb = inode->i_sb;
 731         struct nilfs_inode_info *ii = NILFS_I(inode);
 732         int ret;
 733
 734         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 735                 if (inode->i_data.nrpages)
 736                         truncate_inode_pages(&inode->i_data, 0);
 737                 end_writeback(inode);
 738                 nilfs_clear_inode(inode);
 739                 return;
 740         }
 741         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 742
 743         if (inode->i_data.nrpages)
 744                 truncate_inode_pages(&inode->i_data, 0);
 745
 746         /* TODO: some of the following operations may fail.  */
 747         nilfs_truncate_bmap(ii, 0);
 748         nilfs_mark_inode_dirty(inode);
 749         end_writeback(inode);
 750
 751         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 752         if (!ret)
 753                 atomic_dec(&ii->i_root->inodes_count);
 754
 755         nilfs_clear_inode(inode);
 756
 757         if (IS_SYNC(inode))
 758                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 759         nilfs_transaction_commit(sb);
 760         /* May construct a logical segment and may fail in sync mode.
 761            But delete_inode has no return value. */
 762 }
 763
 764 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 765 {
 766         struct nilfs_transaction_info ti;
 767         struct inode *inode = dentry->d_inode;
 768         struct super_block *sb = inode->i_sb;
 769         int err;
 770
 771         err = inode_change_ok(inode, iattr);
 772         if (err)
 773                 return err;
 774
 775         err = nilfs_transaction_begin(sb, &ti, 0);
 776         if (unlikely(err))
 777                 return err;
 778
 779         if ((iattr->ia_valid & ATTR_SIZE) &&
 780             iattr->ia_size != i_size_read(inode)) {
 781                 inode_dio_wait(inode);
 782
 783                 err = vmtruncate(inode, iattr->ia_size);
 784                 if (unlikely(err))
 785                         goto out_err;
 786         }
 787
 788         setattr_copy(inode, iattr);
 789         mark_inode_dirty(inode);
 790
 791         if (iattr->ia_valid & ATTR_MODE) {
 792                 err = nilfs_acl_chmod(inode);
 793                 if (unlikely(err))
 794                         goto out_err;
 795         }
 796
 797         return nilfs_transaction_commit(sb);
 798
 799 out_err:
 800         nilfs_transaction_abort(sb);
 801         return err;
 802 }
 803
 804 int nilfs_permission(struct inode *inode, int mask)
 805 {
 806         struct nilfs_root *root = NILFS_I(inode)->i_root;
 807         if ((mask & MAY_WRITE) && root &&
 808             root->cno != NILFS_CPTREE_CURRENT_CNO)
 809                 return -EROFS; /* snapshot is not writable */
 810
 811         return generic_permission(inode, mask);
 812 }
 813
 814 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 815 {
 816         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 817         struct nilfs_inode_info *ii = NILFS_I(inode);
 818         int err;
 819
 820         spin_lock(&nilfs->ns_inode_lock);
 821         if (ii->i_bh == NULL) {
 822                 spin_unlock(&nilfs->ns_inode_lock);
 823                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 824                                                   inode->i_ino, pbh);
 825                 if (unlikely(err))
 826                         return err;
 827                 spin_lock(&nilfs->ns_inode_lock);
 828                 if (ii->i_bh == NULL)
 829                         ii->i_bh = *pbh;
 830                 else {
 831                         brelse(*pbh);
 832                         *pbh = ii->i_bh;
 833                 }
 834         } else
 835                 *pbh = ii->i_bh;
 836
 837         get_bh(*pbh);
 838         spin_unlock(&nilfs->ns_inode_lock);
 839         return 0;
 840 }
 841
 842 int nilfs_inode_dirty(struct inode *inode)
 843 {
 844         struct nilfs_inode_info *ii = NILFS_I(inode);
 845         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 846         int ret = 0;
 847
 848         if (!list_empty(&ii->i_dirty)) {
 849                 spin_lock(&nilfs->ns_inode_lock);
 850                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 851                         test_bit(NILFS_I_BUSY, &ii->i_state);
 852                 spin_unlock(&nilfs->ns_inode_lock);
 853         }
 854         return ret;
 855 }
 856
 857 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 858 {
 859         struct nilfs_inode_info *ii = NILFS_I(inode);
 860         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 861
 862         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 863
 864         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 865                 return 0;
 866
 867         spin_lock(&nilfs->ns_inode_lock);
 868         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 869             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 870                 /* Because this routine may race with nilfs_dispose_list(),
 871                    we have to check NILFS_I_QUEUED here, too. */
 872                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 873                         /* This will happen when somebody is freeing
 874                            this inode. */
 875                         nilfs_warning(inode->i_sb, __func__,
 876                                       "cannot get inode (ino=%lu)\n",
 877                                       inode->i_ino);
 878                         spin_unlock(&nilfs->ns_inode_lock);
 879                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 880                                            freeing inode */
 881                 }
 882                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 883                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 884         }
 885         spin_unlock(&nilfs->ns_inode_lock);
 886         return 0;
 887 }
 888
 889 int nilfs_mark_inode_dirty(struct inode *inode)
 890 {
 891         struct buffer_head *ibh;
 892         int err;
 893
 894         err = nilfs_load_inode_block(inode, &ibh);
 895         if (unlikely(err)) {
 896                 nilfs_warning(inode->i_sb, __func__,
 897                               "failed to reget inode block.\n");
 898                 return err;
 899         }
 900         nilfs_update_inode(inode, ibh);
 901         mark_buffer_dirty(ibh);
 902         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 903         brelse(ibh);
 904         return 0;
 905 }
 906
 907 /**
 908  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 909  * @inode: inode of the file to be registered.
 910  *
 911  * nilfs_dirty_inode() loads a inode block containing the specified
 912  * @inode and copies data from a nilfs_inode to a corresponding inode
 913  * entry in the inode block. This operation is excluded from the segment
 914  * construction. This function can be called both as a single operation
 915  * and as a part of indivisible file operations.
 916  */
 917 void nilfs_dirty_inode(struct inode *inode, int flags)
 918 {
 919         struct nilfs_transaction_info ti;
 920         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 921
 922         if (is_bad_inode(inode)) {
 923                 nilfs_warning(inode->i_sb, __func__,
 924                               "tried to mark bad_inode dirty. ignored.\n");
 925                 dump_stack();
 926                 return;
 927         }
 928         if (mdi) {
 929                 nilfs_mdt_mark_dirty(inode);
 930                 return;
 931         }
 932         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 933         nilfs_mark_inode_dirty(inode);
 934         nilfs_transaction_commit(inode->i_sb); /* never fails */
 935 }
 936
 937 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 938                  __u64 start, __u64 len)
 939 {
 940         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 941         __u64 logical = 0, phys = 0, size = 0;
 942         __u32 flags = 0;
 943         loff_t isize;
 944         sector_t blkoff, end_blkoff;
 945         sector_t delalloc_blkoff;
 946         unsigned long delalloc_blklen;
 947         unsigned int blkbits = inode->i_blkbits;
 948         int ret, n;
 949
 950         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 951         if (ret)
 952                 return ret;
 953
 954         mutex_lock(&inode->i_mutex);
 955
 956         isize = i_size_read(inode);
 957
 958         blkoff = start >> blkbits;
 959         end_blkoff = (start + len - 1) >> blkbits;
 960
 961         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
 962                                                         &delalloc_blkoff);
 963
 964         do {
 965                 __u64 blkphy;
 966                 unsigned int maxblocks;
 967
 968                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
 969                         if (size) {
 970                                 /* End of the current extent */
 971                                 ret = fiemap_fill_next_extent(
 972                                         fieinfo, logical, phys, size, flags);
 973                                 if (ret)
 974                                         break;
 975                         }
 976                         if (blkoff > end_blkoff)
 977                                 break;
 978
 979                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
 980                         logical = blkoff << blkbits;
 981                         phys = 0;
 982                         size = delalloc_blklen << blkbits;
 983
 984                         blkoff = delalloc_blkoff + delalloc_blklen;
 985                         delalloc_blklen = nilfs_find_uncommitted_extent(
 986                                 inode, blkoff, &delalloc_blkoff);
 987                         continue;
 988                 }
 989
 990                 /*
 991                  * Limit the number of blocks that we look up so as
 992                  * not to get into the next delayed allocation extent.
 993                  */
 994                 maxblocks = INT_MAX;
 995                 if (delalloc_blklen)
 996                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
 997                                           maxblocks);
 998                 blkphy = 0;
 999
1000                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1001                 n = nilfs_bmap_lookup_contig(
1002                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1003                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1004
1005                 if (n < 0) {
1006                         int past_eof;
1007
1008                         if (unlikely(n != -ENOENT))
1009                                 break; /* error */
1010
1011                         /* HOLE */
1012                         blkoff++;
1013                         past_eof = ((blkoff << blkbits) >= isize);
1014
1015                         if (size) {
1016                                 /* End of the current extent */
1017
1018                                 if (past_eof)
1019                                         flags |= FIEMAP_EXTENT_LAST;
1020
1021                                 ret = fiemap_fill_next_extent(
1022                                         fieinfo, logical, phys, size, flags);
1023                                 if (ret)
1024                                         break;
1025                                 size = 0;
1026                         }
1027                         if (blkoff > end_blkoff || past_eof)
1028                                 break;
1029                 } else {
1030                         if (size) {
1031                                 if (phys && blkphy << blkbits == phys + size) {
1032                                         /* The current extent goes on */
1033                                         size += n << blkbits;
1034                                 } else {
1035                                         /* Terminate the current extent */
1036                                         ret = fiemap_fill_next_extent(
1037                                                 fieinfo, logical, phys, size,
1038                                                 flags);
1039                                         if (ret || blkoff > end_blkoff)
1040                                                 break;
1041
1042                                         /* Start another extent */
1043                                         flags = FIEMAP_EXTENT_MERGED;
1044                                         logical = blkoff << blkbits;
1045                                         phys = blkphy << blkbits;
1046                                         size = n << blkbits;
1047                                 }
1048                         } else {
1049                                 /* Start a new extent */
1050                                 flags = FIEMAP_EXTENT_MERGED;
1051                                 logical = blkoff << blkbits;
1052                                 phys = blkphy << blkbits;
1053                                 size = n << blkbits;
1054                         }
1055                         blkoff += n;
1056                 }
1057                 cond_resched();
1058         } while (true);
1059
1060         /* If ret is 1 then we just hit the end of the extent array */
1061         if (ret == 1)
1062                 ret = 0;
1063
1064         mutex_unlock(&inode->i_mutex);
1065         return ret;
1066 }