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  * Written by Ryusuke Konishi.
  17  *
  18  */
  19
  20 #include <linux/buffer_head.h>
  21 #include <linux/gfp.h>
  22 #include <linux/mpage.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/writeback.h>
  25 #include <linux/uio.h>
  26 #include "nilfs.h"
  27 #include "btnode.h"
  28 #include "segment.h"
  29 #include "page.h"
  30 #include "mdt.h"
  31 #include "cpfile.h"
  32 #include "ifile.h"
  33
  34 /**
  35  * struct nilfs_iget_args - arguments used during comparison between inodes
  36  * @ino: inode number
  37  * @cno: checkpoint number
  38  * @root: pointer on NILFS root object (mounted checkpoint)
  39  * @for_gc: inode for GC flag
  40  */
  41 struct nilfs_iget_args {
  42         u64 ino;
  43         __u64 cno;
  44         struct nilfs_root *root;
  45         int for_gc;
  46 };
  47
  48 static int nilfs_iget_test(struct inode *inode, void *opaque);
  49
  50 void nilfs_inode_add_blocks(struct inode *inode, int n)
  51 {
  52         struct nilfs_root *root = NILFS_I(inode)->i_root;
  53
  54         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  55         if (root)
  56                 atomic64_add(n, &root->blocks_count);
  57 }
  58
  59 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  60 {
  61         struct nilfs_root *root = NILFS_I(inode)->i_root;
  62
  63         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  64         if (root)
  65                 atomic64_sub(n, &root->blocks_count);
  66 }
  67
  68 /**
  69  * nilfs_get_block() - get a file block on the filesystem (callback function)
  70  * @inode - inode struct of the target file
  71  * @blkoff - file block number
  72  * @bh_result - buffer head to be mapped on
  73  * @create - indicate whether allocating the block or not when it has not
  74  *      been allocated yet.
  75  *
  76  * This function does not issue actual read request of the specified data
  77  * block. It is done by VFS.
  78  */
  79 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  80                     struct buffer_head *bh_result, int create)
  81 {
  82         struct nilfs_inode_info *ii = NILFS_I(inode);
  83         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  84         __u64 blknum = 0;
  85         int err = 0, ret;
  86         unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
  87
  88         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  89         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  90         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  91         if (ret >= 0) { /* found */
  92                 map_bh(bh_result, inode->i_sb, blknum);
  93                 if (ret > 0)
  94                         bh_result->b_size = (ret << inode->i_blkbits);
  95                 goto out;
  96         }
  97         /* data block was not found */
  98         if (ret == -ENOENT && create) {
  99                 struct nilfs_transaction_info ti;
 100
 101                 bh_result->b_blocknr = 0;
 102                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 103                 if (unlikely(err))
 104                         goto out;
 105                 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
 106                                         (unsigned long)bh_result);
 107                 if (unlikely(err != 0)) {
 108                         if (err == -EEXIST) {
 109                                 /*
 110                                  * The get_block() function could be called
 111                                  * from multiple callers for an inode.
 112                                  * However, the page having this block must
 113                                  * be locked in this case.
 114                                  */
 115                                 printk(KERN_WARNING
 116                                        "nilfs_get_block: a race condition "
 117                                        "while inserting a data block. "
 118                                        "(inode number=%lu, file block "
 119                                        "offset=%llu)\n",
 120                                        inode->i_ino,
 121                                        (unsigned long long)blkoff);
 122                                 err = 0;
 123                         }
 124                         nilfs_transaction_abort(inode->i_sb);
 125                         goto out;
 126                 }
 127                 nilfs_mark_inode_dirty_sync(inode);
 128                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 129                 /* Error handling should be detailed */
 130                 set_buffer_new(bh_result);
 131                 set_buffer_delay(bh_result);
 132                 map_bh(bh_result, inode->i_sb, 0);
 133                 /* Disk block number must be changed to proper value */
 134
 135         } else if (ret == -ENOENT) {
 136                 /*
 137                  * not found is not error (e.g. hole); must return without
 138                  * the mapped state flag.
 139                  */
 140                 ;
 141         } else {
 142                 err = ret;
 143         }
 144
 145  out:
 146         return err;
 147 }
 148
 149 /**
 150  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 151  * address_space_operations.
 152  * @file - file struct of the file to be read
 153  * @page - the page to be read
 154  */
 155 static int nilfs_readpage(struct file *file, struct page *page)
 156 {
 157         return mpage_readpage(page, nilfs_get_block);
 158 }
 159
 160 /**
 161  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 162  * address_space_operations.
 163  * @file - file struct of the file to be read
 164  * @mapping - address_space struct used for reading multiple pages
 165  * @pages - the pages to be read
 166  * @nr_pages - number of pages to be read
 167  */
 168 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 169                            struct list_head *pages, unsigned int nr_pages)
 170 {
 171         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 172 }
 173
 174 static int nilfs_writepages(struct address_space *mapping,
 175                             struct writeback_control *wbc)
 176 {
 177         struct inode *inode = mapping->host;
 178         int err = 0;
 179
 180         if (inode->i_sb->s_flags & MS_RDONLY) {
 181                 nilfs_clear_dirty_pages(mapping, false);
 182                 return -EROFS;
 183         }
 184
 185         if (wbc->sync_mode == WB_SYNC_ALL)
 186                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 187                                                     wbc->range_start,
 188                                                     wbc->range_end);
 189         return err;
 190 }
 191
 192 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 193 {
 194         struct inode *inode = page->mapping->host;
 195         int err;
 196
 197         if (inode->i_sb->s_flags & MS_RDONLY) {
 198                 /*
 199                  * It means that filesystem was remounted in read-only
 200                  * mode because of error or metadata corruption. But we
 201                  * have dirty pages that try to be flushed in background.
 202                  * So, here we simply discard this dirty page.
 203                  */
 204                 nilfs_clear_dirty_page(page, false);
 205                 unlock_page(page);
 206                 return -EROFS;
 207         }
 208
 209         redirty_page_for_writepage(wbc, page);
 210         unlock_page(page);
 211
 212         if (wbc->sync_mode == WB_SYNC_ALL) {
 213                 err = nilfs_construct_segment(inode->i_sb);
 214                 if (unlikely(err))
 215                         return err;
 216         } else if (wbc->for_reclaim)
 217                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 218
 219         return 0;
 220 }
 221
 222 static int nilfs_set_page_dirty(struct page *page)
 223 {
 224         struct inode *inode = page->mapping->host;
 225         int ret = __set_page_dirty_nobuffers(page);
 226
 227         if (page_has_buffers(page)) {
 228                 unsigned int nr_dirty = 0;
 229                 struct buffer_head *bh, *head;
 230
 231                 /*
 232                  * This page is locked by callers, and no other thread
 233                  * concurrently marks its buffers dirty since they are
 234                  * only dirtied through routines in fs/buffer.c in
 235                  * which call sites of mark_buffer_dirty are protected
 236                  * by page lock.
 237                  */
 238                 bh = head = page_buffers(page);
 239                 do {
 240                         /* Do not mark hole blocks dirty */
 241                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 242                                 continue;
 243
 244                         set_buffer_dirty(bh);
 245                         nr_dirty++;
 246                 } while (bh = bh->b_this_page, bh != head);
 247
 248                 if (nr_dirty)
 249                         nilfs_set_file_dirty(inode, nr_dirty);
 250         } else if (ret) {
 251                 unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 252
 253                 nilfs_set_file_dirty(inode, nr_dirty);
 254         }
 255         return ret;
 256 }
 257
 258 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 259 {
 260         struct inode *inode = mapping->host;
 261
 262         if (to > inode->i_size) {
 263                 truncate_pagecache(inode, inode->i_size);
 264                 nilfs_truncate(inode);
 265         }
 266 }
 267
 268 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 269                              loff_t pos, unsigned len, unsigned flags,
 270                              struct page **pagep, void **fsdata)
 271
 272 {
 273         struct inode *inode = mapping->host;
 274         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 275
 276         if (unlikely(err))
 277                 return err;
 278
 279         err = block_write_begin(mapping, pos, len, flags, pagep,
 280                                 nilfs_get_block);
 281         if (unlikely(err)) {
 282                 nilfs_write_failed(mapping, pos + len);
 283                 nilfs_transaction_abort(inode->i_sb);
 284         }
 285         return err;
 286 }
 287
 288 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 289                            loff_t pos, unsigned len, unsigned copied,
 290                            struct page *page, void *fsdata)
 291 {
 292         struct inode *inode = mapping->host;
 293         unsigned int start = pos & (PAGE_SIZE - 1);
 294         unsigned int nr_dirty;
 295         int err;
 296
 297         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 298                                                   start + copied);
 299         copied = generic_write_end(file, mapping, pos, len, copied, page,
 300                                    fsdata);
 301         nilfs_set_file_dirty(inode, nr_dirty);
 302         err = nilfs_transaction_commit(inode->i_sb);
 303         return err ? : copied;
 304 }
 305
 306 static ssize_t
 307 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 308 {
 309         struct inode *inode = file_inode(iocb->ki_filp);
 310
 311         if (iov_iter_rw(iter) == WRITE)
 312                 return 0;
 313
 314         /* Needs synchronization with the cleaner */
 315         return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
 316 }
 317
 318 const struct address_space_operations nilfs_aops = {
 319         .writepage              = nilfs_writepage,
 320         .readpage               = nilfs_readpage,
 321         .writepages             = nilfs_writepages,
 322         .set_page_dirty         = nilfs_set_page_dirty,
 323         .readpages              = nilfs_readpages,
 324         .write_begin            = nilfs_write_begin,
 325         .write_end              = nilfs_write_end,
 326         /* .releasepage         = nilfs_releasepage, */
 327         .invalidatepage         = block_invalidatepage,
 328         .direct_IO              = nilfs_direct_IO,
 329         .is_partially_uptodate  = block_is_partially_uptodate,
 330 };
 331
 332 static int nilfs_insert_inode_locked(struct inode *inode,
 333                                      struct nilfs_root *root,
 334                                      unsigned long ino)
 335 {
 336         struct nilfs_iget_args args = {
 337                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 338         };
 339
 340         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 341 }
 342
 343 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 344 {
 345         struct super_block *sb = dir->i_sb;
 346         struct the_nilfs *nilfs = sb->s_fs_info;
 347         struct inode *inode;
 348         struct nilfs_inode_info *ii;
 349         struct nilfs_root *root;
 350         int err = -ENOMEM;
 351         ino_t ino;
 352
 353         inode = new_inode(sb);
 354         if (unlikely(!inode))
 355                 goto failed;
 356
 357         mapping_set_gfp_mask(inode->i_mapping,
 358                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 359
 360         root = NILFS_I(dir)->i_root;
 361         ii = NILFS_I(inode);
 362         ii->i_state = 1 << NILFS_I_NEW;
 363         ii->i_root = root;
 364
 365         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 366         if (unlikely(err))
 367                 goto failed_ifile_create_inode;
 368         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 369
 370         atomic64_inc(&root->inodes_count);
 371         inode_init_owner(inode, dir, mode);
 372         inode->i_ino = ino;
 373         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 374
 375         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 376                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 377                 if (err < 0)
 378                         goto failed_after_creation;
 379
 380                 set_bit(NILFS_I_BMAP, &ii->i_state);
 381                 /* No lock is needed; iget() ensures it. */
 382         }
 383
 384         ii->i_flags = nilfs_mask_flags(
 385                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 386
 387         /* ii->i_file_acl = 0; */
 388         /* ii->i_dir_acl = 0; */
 389         ii->i_dir_start_lookup = 0;
 390         nilfs_set_inode_flags(inode);
 391         spin_lock(&nilfs->ns_next_gen_lock);
 392         inode->i_generation = nilfs->ns_next_generation++;
 393         spin_unlock(&nilfs->ns_next_gen_lock);
 394         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 395                 err = -EIO;
 396                 goto failed_after_creation;
 397         }
 398
 399         err = nilfs_init_acl(inode, dir);
 400         if (unlikely(err))
 401                 /*
 402                  * Never occur.  When supporting nilfs_init_acl(),
 403                  * proper cancellation of above jobs should be considered.
 404                  */
 405                 goto failed_after_creation;
 406
 407         return inode;
 408
 409  failed_after_creation:
 410         clear_nlink(inode);
 411         unlock_new_inode(inode);
 412         iput(inode);  /*
 413                        * raw_inode will be deleted through
 414                        * nilfs_evict_inode().
 415                        */
 416         goto failed;
 417
 418  failed_ifile_create_inode:
 419         make_bad_inode(inode);
 420         iput(inode);
 421  failed:
 422         return ERR_PTR(err);
 423 }
 424
 425 void nilfs_set_inode_flags(struct inode *inode)
 426 {
 427         unsigned int flags = NILFS_I(inode)->i_flags;
 428         unsigned int new_fl = 0;
 429
 430         if (flags & FS_SYNC_FL)
 431                 new_fl |= S_SYNC;
 432         if (flags & FS_APPEND_FL)
 433                 new_fl |= S_APPEND;
 434         if (flags & FS_IMMUTABLE_FL)
 435                 new_fl |= S_IMMUTABLE;
 436         if (flags & FS_NOATIME_FL)
 437                 new_fl |= S_NOATIME;
 438         if (flags & FS_DIRSYNC_FL)
 439                 new_fl |= S_DIRSYNC;
 440         inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
 441                         S_NOATIME | S_DIRSYNC);
 442 }
 443
 444 int nilfs_read_inode_common(struct inode *inode,
 445                             struct nilfs_inode *raw_inode)
 446 {
 447         struct nilfs_inode_info *ii = NILFS_I(inode);
 448         int err;
 449
 450         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 451         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 452         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 453         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 454         inode->i_size = le64_to_cpu(raw_inode->i_size);
 455         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 456         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 457         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 458         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 459         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 460         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 461         if (inode->i_nlink == 0)
 462                 return -ESTALE; /* this inode is deleted */
 463
 464         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 465         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 466 #if 0
 467         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 468         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 469                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 470 #endif
 471         ii->i_dir_start_lookup = 0;
 472         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 473
 474         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 475             S_ISLNK(inode->i_mode)) {
 476                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 477                 if (err < 0)
 478                         return err;
 479                 set_bit(NILFS_I_BMAP, &ii->i_state);
 480                 /* No lock is needed; iget() ensures it. */
 481         }
 482         return 0;
 483 }
 484
 485 static int __nilfs_read_inode(struct super_block *sb,
 486                               struct nilfs_root *root, unsigned long ino,
 487                               struct inode *inode)
 488 {
 489         struct the_nilfs *nilfs = sb->s_fs_info;
 490         struct buffer_head *bh;
 491         struct nilfs_inode *raw_inode;
 492         int err;
 493
 494         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 495         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 496         if (unlikely(err))
 497                 goto bad_inode;
 498
 499         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 500
 501         err = nilfs_read_inode_common(inode, raw_inode);
 502         if (err)
 503                 goto failed_unmap;
 504
 505         if (S_ISREG(inode->i_mode)) {
 506                 inode->i_op = &nilfs_file_inode_operations;
 507                 inode->i_fop = &nilfs_file_operations;
 508                 inode->i_mapping->a_ops = &nilfs_aops;
 509         } else if (S_ISDIR(inode->i_mode)) {
 510                 inode->i_op = &nilfs_dir_inode_operations;
 511                 inode->i_fop = &nilfs_dir_operations;
 512                 inode->i_mapping->a_ops = &nilfs_aops;
 513         } else if (S_ISLNK(inode->i_mode)) {
 514                 inode->i_op = &nilfs_symlink_inode_operations;
 515                 inode_nohighmem(inode);
 516                 inode->i_mapping->a_ops = &nilfs_aops;
 517         } else {
 518                 inode->i_op = &nilfs_special_inode_operations;
 519                 init_special_inode(
 520                         inode, inode->i_mode,
 521                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 522         }
 523         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 524         brelse(bh);
 525         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 526         nilfs_set_inode_flags(inode);
 527         mapping_set_gfp_mask(inode->i_mapping,
 528                            mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
 529         return 0;
 530
 531  failed_unmap:
 532         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 533         brelse(bh);
 534
 535  bad_inode:
 536         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 537         return err;
 538 }
 539
 540 static int nilfs_iget_test(struct inode *inode, void *opaque)
 541 {
 542         struct nilfs_iget_args *args = opaque;
 543         struct nilfs_inode_info *ii;
 544
 545         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 546                 return 0;
 547
 548         ii = NILFS_I(inode);
 549         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 550                 return !args->for_gc;
 551
 552         return args->for_gc && args->cno == ii->i_cno;
 553 }
 554
 555 static int nilfs_iget_set(struct inode *inode, void *opaque)
 556 {
 557         struct nilfs_iget_args *args = opaque;
 558
 559         inode->i_ino = args->ino;
 560         if (args->for_gc) {
 561                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 562                 NILFS_I(inode)->i_cno = args->cno;
 563                 NILFS_I(inode)->i_root = NULL;
 564         } else {
 565                 if (args->root && args->ino == NILFS_ROOT_INO)
 566                         nilfs_get_root(args->root);
 567                 NILFS_I(inode)->i_root = args->root;
 568         }
 569         return 0;
 570 }
 571
 572 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 573                             unsigned long ino)
 574 {
 575         struct nilfs_iget_args args = {
 576                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 577         };
 578
 579         return ilookup5(sb, ino, nilfs_iget_test, &args);
 580 }
 581
 582 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 583                                 unsigned long ino)
 584 {
 585         struct nilfs_iget_args args = {
 586                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 587         };
 588
 589         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 590 }
 591
 592 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 593                          unsigned long ino)
 594 {
 595         struct inode *inode;
 596         int err;
 597
 598         inode = nilfs_iget_locked(sb, root, ino);
 599         if (unlikely(!inode))
 600                 return ERR_PTR(-ENOMEM);
 601         if (!(inode->i_state & I_NEW))
 602                 return inode;
 603
 604         err = __nilfs_read_inode(sb, root, ino, inode);
 605         if (unlikely(err)) {
 606                 iget_failed(inode);
 607                 return ERR_PTR(err);
 608         }
 609         unlock_new_inode(inode);
 610         return inode;
 611 }
 612
 613 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 614                                 __u64 cno)
 615 {
 616         struct nilfs_iget_args args = {
 617                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 618         };
 619         struct inode *inode;
 620         int err;
 621
 622         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 623         if (unlikely(!inode))
 624                 return ERR_PTR(-ENOMEM);
 625         if (!(inode->i_state & I_NEW))
 626                 return inode;
 627
 628         err = nilfs_init_gcinode(inode);
 629         if (unlikely(err)) {
 630                 iget_failed(inode);
 631                 return ERR_PTR(err);
 632         }
 633         unlock_new_inode(inode);
 634         return inode;
 635 }
 636
 637 void nilfs_write_inode_common(struct inode *inode,
 638                               struct nilfs_inode *raw_inode, int has_bmap)
 639 {
 640         struct nilfs_inode_info *ii = NILFS_I(inode);
 641
 642         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 643         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 644         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 645         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 646         raw_inode->i_size = cpu_to_le64(inode->i_size);
 647         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 648         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 649         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 650         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 651         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 652
 653         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 654         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 655
 656         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 657                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 658
 659                 /* zero-fill unused portion in the case of super root block */
 660                 raw_inode->i_xattr = 0;
 661                 raw_inode->i_pad = 0;
 662                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 663                        nilfs->ns_inode_size - sizeof(*raw_inode));
 664         }
 665
 666         if (has_bmap)
 667                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 668         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 669                 raw_inode->i_device_code =
 670                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 671         /*
 672          * When extending inode, nilfs->ns_inode_size should be checked
 673          * for substitutions of appended fields.
 674          */
 675 }
 676
 677 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 678 {
 679         ino_t ino = inode->i_ino;
 680         struct nilfs_inode_info *ii = NILFS_I(inode);
 681         struct inode *ifile = ii->i_root->ifile;
 682         struct nilfs_inode *raw_inode;
 683
 684         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 685
 686         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 687                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 688         if (flags & I_DIRTY_DATASYNC)
 689                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 690
 691         nilfs_write_inode_common(inode, raw_inode, 0);
 692                 /*
 693                  * XXX: call with has_bmap = 0 is a workaround to avoid
 694                  * deadlock of bmap.  This delays update of i_bmap to just
 695                  * before writing.
 696                  */
 697
 698         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 699 }
 700
 701 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 702
 703 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 704                                 unsigned long from)
 705 {
 706         __u64 b;
 707         int ret;
 708
 709         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 710                 return;
 711 repeat:
 712         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 713         if (ret == -ENOENT)
 714                 return;
 715         else if (ret < 0)
 716                 goto failed;
 717
 718         if (b < from)
 719                 return;
 720
 721         b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 722         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 723         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 724         if (!ret || (ret == -ENOMEM &&
 725                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 726                 goto repeat;
 727
 728 failed:
 729         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 730                       "failed to truncate bmap (ino=%lu, err=%d)",
 731                       ii->vfs_inode.i_ino, ret);
 732 }
 733
 734 void nilfs_truncate(struct inode *inode)
 735 {
 736         unsigned long blkoff;
 737         unsigned int blocksize;
 738         struct nilfs_transaction_info ti;
 739         struct super_block *sb = inode->i_sb;
 740         struct nilfs_inode_info *ii = NILFS_I(inode);
 741
 742         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 743                 return;
 744         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 745                 return;
 746
 747         blocksize = sb->s_blocksize;
 748         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 749         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 750
 751         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 752
 753         nilfs_truncate_bmap(ii, blkoff);
 754
 755         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 756         if (IS_SYNC(inode))
 757                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 758
 759         nilfs_mark_inode_dirty(inode);
 760         nilfs_set_file_dirty(inode, 0);
 761         nilfs_transaction_commit(sb);
 762         /*
 763          * May construct a logical segment and may fail in sync mode.
 764          * But truncate has no return value.
 765          */
 766 }
 767
 768 static void nilfs_clear_inode(struct inode *inode)
 769 {
 770         struct nilfs_inode_info *ii = NILFS_I(inode);
 771
 772         /*
 773          * Free resources allocated in nilfs_read_inode(), here.
 774          */
 775         BUG_ON(!list_empty(&ii->i_dirty));
 776         brelse(ii->i_bh);
 777         ii->i_bh = NULL;
 778
 779         if (nilfs_is_metadata_file_inode(inode))
 780                 nilfs_mdt_clear(inode);
 781
 782         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 783                 nilfs_bmap_clear(ii->i_bmap);
 784
 785         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 786
 787         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 788                 nilfs_put_root(ii->i_root);
 789 }
 790
 791 void nilfs_evict_inode(struct inode *inode)
 792 {
 793         struct nilfs_transaction_info ti;
 794         struct super_block *sb = inode->i_sb;
 795         struct nilfs_inode_info *ii = NILFS_I(inode);
 796         int ret;
 797
 798         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 799                 truncate_inode_pages_final(&inode->i_data);
 800                 clear_inode(inode);
 801                 nilfs_clear_inode(inode);
 802                 return;
 803         }
 804         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 805
 806         truncate_inode_pages_final(&inode->i_data);
 807
 808         /* TODO: some of the following operations may fail.  */
 809         nilfs_truncate_bmap(ii, 0);
 810         nilfs_mark_inode_dirty(inode);
 811         clear_inode(inode);
 812
 813         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 814         if (!ret)
 815                 atomic64_dec(&ii->i_root->inodes_count);
 816
 817         nilfs_clear_inode(inode);
 818
 819         if (IS_SYNC(inode))
 820                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 821         nilfs_transaction_commit(sb);
 822         /*
 823          * May construct a logical segment and may fail in sync mode.
 824          * But delete_inode has no return value.
 825          */
 826 }
 827
 828 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 829 {
 830         struct nilfs_transaction_info ti;
 831         struct inode *inode = d_inode(dentry);
 832         struct super_block *sb = inode->i_sb;
 833         int err;
 834
 835         err = inode_change_ok(inode, iattr);
 836         if (err)
 837                 return err;
 838
 839         err = nilfs_transaction_begin(sb, &ti, 0);
 840         if (unlikely(err))
 841                 return err;
 842
 843         if ((iattr->ia_valid & ATTR_SIZE) &&
 844             iattr->ia_size != i_size_read(inode)) {
 845                 inode_dio_wait(inode);
 846                 truncate_setsize(inode, iattr->ia_size);
 847                 nilfs_truncate(inode);
 848         }
 849
 850         setattr_copy(inode, iattr);
 851         mark_inode_dirty(inode);
 852
 853         if (iattr->ia_valid & ATTR_MODE) {
 854                 err = nilfs_acl_chmod(inode);
 855                 if (unlikely(err))
 856                         goto out_err;
 857         }
 858
 859         return nilfs_transaction_commit(sb);
 860
 861 out_err:
 862         nilfs_transaction_abort(sb);
 863         return err;
 864 }
 865
 866 int nilfs_permission(struct inode *inode, int mask)
 867 {
 868         struct nilfs_root *root = NILFS_I(inode)->i_root;
 869
 870         if ((mask & MAY_WRITE) && root &&
 871             root->cno != NILFS_CPTREE_CURRENT_CNO)
 872                 return -EROFS; /* snapshot is not writable */
 873
 874         return generic_permission(inode, mask);
 875 }
 876
 877 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 878 {
 879         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 880         struct nilfs_inode_info *ii = NILFS_I(inode);
 881         int err;
 882
 883         spin_lock(&nilfs->ns_inode_lock);
 884         if (ii->i_bh == NULL) {
 885                 spin_unlock(&nilfs->ns_inode_lock);
 886                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 887                                                   inode->i_ino, pbh);
 888                 if (unlikely(err))
 889                         return err;
 890                 spin_lock(&nilfs->ns_inode_lock);
 891                 if (ii->i_bh == NULL)
 892                         ii->i_bh = *pbh;
 893                 else {
 894                         brelse(*pbh);
 895                         *pbh = ii->i_bh;
 896                 }
 897         } else
 898                 *pbh = ii->i_bh;
 899
 900         get_bh(*pbh);
 901         spin_unlock(&nilfs->ns_inode_lock);
 902         return 0;
 903 }
 904
 905 int nilfs_inode_dirty(struct inode *inode)
 906 {
 907         struct nilfs_inode_info *ii = NILFS_I(inode);
 908         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 909         int ret = 0;
 910
 911         if (!list_empty(&ii->i_dirty)) {
 912                 spin_lock(&nilfs->ns_inode_lock);
 913                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 914                         test_bit(NILFS_I_BUSY, &ii->i_state);
 915                 spin_unlock(&nilfs->ns_inode_lock);
 916         }
 917         return ret;
 918 }
 919
 920 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
 921 {
 922         struct nilfs_inode_info *ii = NILFS_I(inode);
 923         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 924
 925         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 926
 927         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 928                 return 0;
 929
 930         spin_lock(&nilfs->ns_inode_lock);
 931         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 932             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 933                 /*
 934                  * Because this routine may race with nilfs_dispose_list(),
 935                  * we have to check NILFS_I_QUEUED here, too.
 936                  */
 937                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 938                         /*
 939                          * This will happen when somebody is freeing
 940                          * this inode.
 941                          */
 942                         nilfs_warning(inode->i_sb, __func__,
 943                                       "cannot get inode (ino=%lu)",
 944                                       inode->i_ino);
 945                         spin_unlock(&nilfs->ns_inode_lock);
 946                         return -EINVAL; /*
 947                                          * NILFS_I_DIRTY may remain for
 948                                          * freeing inode.
 949                                          */
 950                 }
 951                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 952                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 953         }
 954         spin_unlock(&nilfs->ns_inode_lock);
 955         return 0;
 956 }
 957
 958 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
 959 {
 960         struct buffer_head *ibh;
 961         int err;
 962
 963         err = nilfs_load_inode_block(inode, &ibh);
 964         if (unlikely(err)) {
 965                 nilfs_warning(inode->i_sb, __func__,
 966                               "failed to reget inode block.");
 967                 return err;
 968         }
 969         nilfs_update_inode(inode, ibh, flags);
 970         mark_buffer_dirty(ibh);
 971         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 972         brelse(ibh);
 973         return 0;
 974 }
 975
 976 /**
 977  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 978  * @inode: inode of the file to be registered.
 979  *
 980  * nilfs_dirty_inode() loads a inode block containing the specified
 981  * @inode and copies data from a nilfs_inode to a corresponding inode
 982  * entry in the inode block. This operation is excluded from the segment
 983  * construction. This function can be called both as a single operation
 984  * and as a part of indivisible file operations.
 985  */
 986 void nilfs_dirty_inode(struct inode *inode, int flags)
 987 {
 988         struct nilfs_transaction_info ti;
 989         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 990
 991         if (is_bad_inode(inode)) {
 992                 nilfs_warning(inode->i_sb, __func__,
 993                               "tried to mark bad_inode dirty. ignored.");
 994                 dump_stack();
 995                 return;
 996         }
 997         if (mdi) {
 998                 nilfs_mdt_mark_dirty(inode);
 999                 return;
1000         }
1001         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1002         __nilfs_mark_inode_dirty(inode, flags);
1003         nilfs_transaction_commit(inode->i_sb); /* never fails */
1004 }
1005
1006 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1007                  __u64 start, __u64 len)
1008 {
1009         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1010         __u64 logical = 0, phys = 0, size = 0;
1011         __u32 flags = 0;
1012         loff_t isize;
1013         sector_t blkoff, end_blkoff;
1014         sector_t delalloc_blkoff;
1015         unsigned long delalloc_blklen;
1016         unsigned int blkbits = inode->i_blkbits;
1017         int ret, n;
1018
1019         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1020         if (ret)
1021                 return ret;
1022
1023         inode_lock(inode);
1024
1025         isize = i_size_read(inode);
1026
1027         blkoff = start >> blkbits;
1028         end_blkoff = (start + len - 1) >> blkbits;
1029
1030         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1031                                                         &delalloc_blkoff);
1032
1033         do {
1034                 __u64 blkphy;
1035                 unsigned int maxblocks;
1036
1037                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1038                         if (size) {
1039                                 /* End of the current extent */
1040                                 ret = fiemap_fill_next_extent(
1041                                         fieinfo, logical, phys, size, flags);
1042                                 if (ret)
1043                                         break;
1044                         }
1045                         if (blkoff > end_blkoff)
1046                                 break;
1047
1048                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1049                         logical = blkoff << blkbits;
1050                         phys = 0;
1051                         size = delalloc_blklen << blkbits;
1052
1053                         blkoff = delalloc_blkoff + delalloc_blklen;
1054                         delalloc_blklen = nilfs_find_uncommitted_extent(
1055                                 inode, blkoff, &delalloc_blkoff);
1056                         continue;
1057                 }
1058
1059                 /*
1060                  * Limit the number of blocks that we look up so as
1061                  * not to get into the next delayed allocation extent.
1062                  */
1063                 maxblocks = INT_MAX;
1064                 if (delalloc_blklen)
1065                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1066                                           maxblocks);
1067                 blkphy = 0;
1068
1069                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1070                 n = nilfs_bmap_lookup_contig(
1071                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1072                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1073
1074                 if (n < 0) {
1075                         int past_eof;
1076
1077                         if (unlikely(n != -ENOENT))
1078                                 break; /* error */
1079
1080                         /* HOLE */
1081                         blkoff++;
1082                         past_eof = ((blkoff << blkbits) >= isize);
1083
1084                         if (size) {
1085                                 /* End of the current extent */
1086
1087                                 if (past_eof)
1088                                         flags |= FIEMAP_EXTENT_LAST;
1089
1090                                 ret = fiemap_fill_next_extent(
1091                                         fieinfo, logical, phys, size, flags);
1092                                 if (ret)
1093                                         break;
1094                                 size = 0;
1095                         }
1096                         if (blkoff > end_blkoff || past_eof)
1097                                 break;
1098                 } else {
1099                         if (size) {
1100                                 if (phys && blkphy << blkbits == phys + size) {
1101                                         /* The current extent goes on */
1102                                         size += n << blkbits;
1103                                 } else {
1104                                         /* Terminate the current extent */
1105                                         ret = fiemap_fill_next_extent(
1106                                                 fieinfo, logical, phys, size,
1107                                                 flags);
1108                                         if (ret || blkoff > end_blkoff)
1109                                                 break;
1110
1111                                         /* Start another extent */
1112                                         flags = FIEMAP_EXTENT_MERGED;
1113                                         logical = blkoff << blkbits;
1114                                         phys = blkphy << blkbits;
1115                                         size = n << blkbits;
1116                                 }
1117                         } else {
1118                                 /* Start a new extent */
1119                                 flags = FIEMAP_EXTENT_MERGED;
1120                                 logical = blkoff << blkbits;
1121                                 phys = blkphy << blkbits;
1122                                 size = n << blkbits;
1123                         }
1124                         blkoff += n;
1125                 }
1126                 cond_resched();
1127         } while (true);
1128
1129         /* If ret is 1 then we just hit the end of the extent array */
1130         if (ret == 1)
1131                 ret = 0;
1132
1133         inode_unlock(inode);
1134         return ret;
1135 }