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/pagemap.h>
  28 #include <linux/writeback.h>
  29 #include <linux/uio.h>
  30 #include "nilfs.h"
  31 #include "btnode.h"
  32 #include "segment.h"
  33 #include "page.h"
  34 #include "mdt.h"
  35 #include "cpfile.h"
  36 #include "ifile.h"
  37
  38 /**
  39  * struct nilfs_iget_args - arguments used during comparison between inodes
  40  * @ino: inode number
  41  * @cno: checkpoint number
  42  * @root: pointer on NILFS root object (mounted checkpoint)
  43  * @for_gc: inode for GC flag
  44  */
  45 struct nilfs_iget_args {
  46         u64 ino;
  47         __u64 cno;
  48         struct nilfs_root *root;
  49         int for_gc;
  50 };
  51
  52 static int nilfs_iget_test(struct inode *inode, void *opaque);
  53
  54 void nilfs_inode_add_blocks(struct inode *inode, int n)
  55 {
  56         struct nilfs_root *root = NILFS_I(inode)->i_root;
  57
  58         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  59         if (root)
  60                 atomic64_add(n, &root->blocks_count);
  61 }
  62
  63 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  64 {
  65         struct nilfs_root *root = NILFS_I(inode)->i_root;
  66
  67         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  68         if (root)
  69                 atomic64_sub(n, &root->blocks_count);
  70 }
  71
  72 /**
  73  * nilfs_get_block() - get a file block on the filesystem (callback function)
  74  * @inode - inode struct of the target file
  75  * @blkoff - file block number
  76  * @bh_result - buffer head to be mapped on
  77  * @create - indicate whether allocating the block or not when it has not
  78  *      been allocated yet.
  79  *
  80  * This function does not issue actual read request of the specified data
  81  * block. It is done by VFS.
  82  */
  83 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  84                     struct buffer_head *bh_result, int create)
  85 {
  86         struct nilfs_inode_info *ii = NILFS_I(inode);
  87         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  88         __u64 blknum = 0;
  89         int err = 0, ret;
  90         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  91
  92         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  93         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  94         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  95         if (ret >= 0) { /* found */
  96                 map_bh(bh_result, inode->i_sb, blknum);
  97                 if (ret > 0)
  98                         bh_result->b_size = (ret << inode->i_blkbits);
  99                 goto out;
 100         }
 101         /* data block was not found */
 102         if (ret == -ENOENT && create) {
 103                 struct nilfs_transaction_info ti;
 104
 105                 bh_result->b_blocknr = 0;
 106                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 107                 if (unlikely(err))
 108                         goto out;
 109                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 110                                         (unsigned long)bh_result);
 111                 if (unlikely(err != 0)) {
 112                         if (err == -EEXIST) {
 113                                 /*
 114                                  * The get_block() function could be called
 115                                  * from multiple callers for an inode.
 116                                  * However, the page having this block must
 117                                  * be locked in this case.
 118                                  */
 119                                 printk(KERN_WARNING
 120                                        "nilfs_get_block: a race condition "
 121                                        "while inserting a data block. "
 122                                        "(inode number=%lu, file block "
 123                                        "offset=%llu)\n",
 124                                        inode->i_ino,
 125                                        (unsigned long long)blkoff);
 126                                 err = 0;
 127                         }
 128                         nilfs_transaction_abort(inode->i_sb);
 129                         goto out;
 130                 }
 131                 nilfs_mark_inode_dirty_sync(inode);
 132                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 133                 /* Error handling should be detailed */
 134                 set_buffer_new(bh_result);
 135                 set_buffer_delay(bh_result);
 136                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 137                                                       to proper value */
 138         } else if (ret == -ENOENT) {
 139                 /* not found is not error (e.g. hole); must return without
 140                    the mapped state flag. */
 141                 ;
 142         } else {
 143                 err = ret;
 144         }
 145
 146  out:
 147         return err;
 148 }
 149
 150 /**
 151  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 152  * address_space_operations.
 153  * @file - file struct of the file to be read
 154  * @page - the page to be read
 155  */
 156 static int nilfs_readpage(struct file *file, struct page *page)
 157 {
 158         return mpage_readpage(page, nilfs_get_block);
 159 }
 160
 161 /**
 162  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 163  * address_space_operations.
 164  * @file - file struct of the file to be read
 165  * @mapping - address_space struct used for reading multiple pages
 166  * @pages - the pages to be read
 167  * @nr_pages - number of pages to be read
 168  */
 169 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 170                            struct list_head *pages, unsigned nr_pages)
 171 {
 172         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 173 }
 174
 175 static int nilfs_writepages(struct address_space *mapping,
 176                             struct writeback_control *wbc)
 177 {
 178         struct inode *inode = mapping->host;
 179         int err = 0;
 180
 181         if (inode->i_sb->s_flags & MS_RDONLY) {
 182                 nilfs_clear_dirty_pages(mapping, false);
 183                 return -EROFS;
 184         }
 185
 186         if (wbc->sync_mode == WB_SYNC_ALL)
 187                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 188                                                     wbc->range_start,
 189                                                     wbc->range_end);
 190         return err;
 191 }
 192
 193 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 194 {
 195         struct inode *inode = page->mapping->host;
 196         int err;
 197
 198         if (inode->i_sb->s_flags & MS_RDONLY) {
 199                 /*
 200                  * It means that filesystem was remounted in read-only
 201                  * mode because of error or metadata corruption. But we
 202                  * have dirty pages that try to be flushed in background.
 203                  * So, here we simply discard this dirty page.
 204                  */
 205                 nilfs_clear_dirty_page(page, false);
 206                 unlock_page(page);
 207                 return -EROFS;
 208         }
 209
 210         redirty_page_for_writepage(wbc, page);
 211         unlock_page(page);
 212
 213         if (wbc->sync_mode == WB_SYNC_ALL) {
 214                 err = nilfs_construct_segment(inode->i_sb);
 215                 if (unlikely(err))
 216                         return err;
 217         } else if (wbc->for_reclaim)
 218                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 219
 220         return 0;
 221 }
 222
 223 static int nilfs_set_page_dirty(struct page *page)
 224 {
 225         struct inode *inode = page->mapping->host;
 226         int ret = __set_page_dirty_nobuffers(page);
 227
 228         if (page_has_buffers(page)) {
 229                 unsigned nr_dirty = 0;
 230                 struct buffer_head *bh, *head;
 231
 232                 /*
 233                  * This page is locked by callers, and no other thread
 234                  * concurrently marks its buffers dirty since they are
 235                  * only dirtied through routines in fs/buffer.c in
 236                  * which call sites of mark_buffer_dirty are protected
 237                  * by page lock.
 238                  */
 239                 bh = head = page_buffers(page);
 240                 do {
 241                         /* Do not mark hole blocks dirty */
 242                         if (buffer_dirty(bh) || !buffer_mapped(bh))
 243                                 continue;
 244
 245                         set_buffer_dirty(bh);
 246                         nr_dirty++;
 247                 } while (bh = bh->b_this_page, bh != head);
 248
 249                 if (nr_dirty)
 250                         nilfs_set_file_dirty(inode, nr_dirty);
 251         } else if (ret) {
 252                 unsigned nr_dirty = 1 << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 253
 254                 nilfs_set_file_dirty(inode, nr_dirty);
 255         }
 256         return ret;
 257 }
 258
 259 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 260 {
 261         struct inode *inode = mapping->host;
 262
 263         if (to > inode->i_size) {
 264                 truncate_pagecache(inode, inode->i_size);
 265                 nilfs_truncate(inode);
 266         }
 267 }
 268
 269 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 270                              loff_t pos, unsigned len, unsigned flags,
 271                              struct page **pagep, void **fsdata)
 272
 273 {
 274         struct inode *inode = mapping->host;
 275         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 276
 277         if (unlikely(err))
 278                 return err;
 279
 280         err = block_write_begin(mapping, pos, len, flags, pagep,
 281                                 nilfs_get_block);
 282         if (unlikely(err)) {
 283                 nilfs_write_failed(mapping, pos + len);
 284                 nilfs_transaction_abort(inode->i_sb);
 285         }
 286         return err;
 287 }
 288
 289 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 290                            loff_t pos, unsigned len, unsigned copied,
 291                            struct page *page, void *fsdata)
 292 {
 293         struct inode *inode = mapping->host;
 294         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 295         unsigned nr_dirty;
 296         int err;
 297
 298         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 299                                                   start + copied);
 300         copied = generic_write_end(file, mapping, pos, len, copied, page,
 301                                    fsdata);
 302         nilfs_set_file_dirty(inode, nr_dirty);
 303         err = nilfs_transaction_commit(inode->i_sb);
 304         return err ? : copied;
 305 }
 306
 307 static ssize_t
 308 nilfs_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
 309                 loff_t offset)
 310 {
 311         struct file *file = iocb->ki_filp;
 312         struct address_space *mapping = file->f_mapping;
 313         struct inode *inode = file->f_mapping->host;
 314         size_t count = iov_iter_count(iter);
 315         ssize_t size;
 316
 317         if (rw == WRITE)
 318                 return 0;
 319
 320         /* Needs synchronization with the cleaner */
 321         size = blockdev_direct_IO(rw, iocb, inode, iter, offset,
 322                                   nilfs_get_block);
 323
 324         /*
 325          * In case of error extending write may have instantiated a few
 326          * blocks outside i_size. Trim these off again.
 327          */
 328         if (unlikely((rw & WRITE) && size < 0)) {
 329                 loff_t isize = i_size_read(inode);
 330                 loff_t end = offset + count;
 331
 332                 if (end > isize)
 333                         nilfs_write_failed(mapping, end);
 334         }
 335
 336         return size;
 337 }
 338
 339 const struct address_space_operations nilfs_aops = {
 340         .writepage              = nilfs_writepage,
 341         .readpage               = nilfs_readpage,
 342         .writepages             = nilfs_writepages,
 343         .set_page_dirty         = nilfs_set_page_dirty,
 344         .readpages              = nilfs_readpages,
 345         .write_begin            = nilfs_write_begin,
 346         .write_end              = nilfs_write_end,
 347         /* .releasepage         = nilfs_releasepage, */
 348         .invalidatepage         = block_invalidatepage,
 349         .direct_IO              = nilfs_direct_IO,
 350         .is_partially_uptodate  = block_is_partially_uptodate,
 351 };
 352
 353 static int nilfs_insert_inode_locked(struct inode *inode,
 354                                      struct nilfs_root *root,
 355                                      unsigned long ino)
 356 {
 357         struct nilfs_iget_args args = {
 358                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 359         };
 360
 361         return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
 362 }
 363
 364 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 365 {
 366         struct super_block *sb = dir->i_sb;
 367         struct the_nilfs *nilfs = sb->s_fs_info;
 368         struct inode *inode;
 369         struct nilfs_inode_info *ii;
 370         struct nilfs_root *root;
 371         int err = -ENOMEM;
 372         ino_t ino;
 373
 374         inode = new_inode(sb);
 375         if (unlikely(!inode))
 376                 goto failed;
 377
 378         mapping_set_gfp_mask(inode->i_mapping,
 379                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 380
 381         root = NILFS_I(dir)->i_root;
 382         ii = NILFS_I(inode);
 383         ii->i_state = 1 << NILFS_I_NEW;
 384         ii->i_root = root;
 385
 386         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 387         if (unlikely(err))
 388                 goto failed_ifile_create_inode;
 389         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 390
 391         atomic64_inc(&root->inodes_count);
 392         inode_init_owner(inode, dir, mode);
 393         inode->i_ino = ino;
 394         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 395
 396         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 397                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 398                 if (err < 0)
 399                         goto failed_after_creation;
 400
 401                 set_bit(NILFS_I_BMAP, &ii->i_state);
 402                 /* No lock is needed; iget() ensures it. */
 403         }
 404
 405         ii->i_flags = nilfs_mask_flags(
 406                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 407
 408         /* ii->i_file_acl = 0; */
 409         /* ii->i_dir_acl = 0; */
 410         ii->i_dir_start_lookup = 0;
 411         nilfs_set_inode_flags(inode);
 412         spin_lock(&nilfs->ns_next_gen_lock);
 413         inode->i_generation = nilfs->ns_next_generation++;
 414         spin_unlock(&nilfs->ns_next_gen_lock);
 415         if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
 416                 err = -EIO;
 417                 goto failed_after_creation;
 418         }
 419
 420         err = nilfs_init_acl(inode, dir);
 421         if (unlikely(err))
 422                 goto failed_after_creation; /* never occur. When supporting
 423                                     nilfs_init_acl(), proper cancellation of
 424                                     above jobs should be considered */
 425
 426         return inode;
 427
 428  failed_after_creation:
 429         clear_nlink(inode);
 430         unlock_new_inode(inode);
 431         iput(inode);  /* raw_inode will be deleted through
 432                          nilfs_evict_inode() */
 433         goto failed;
 434
 435  failed_ifile_create_inode:
 436         make_bad_inode(inode);
 437         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 438                          called */
 439  failed:
 440         return ERR_PTR(err);
 441 }
 442
 443 void nilfs_set_inode_flags(struct inode *inode)
 444 {
 445         unsigned int flags = NILFS_I(inode)->i_flags;
 446
 447         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 448                             S_DIRSYNC);
 449         if (flags & FS_SYNC_FL)
 450                 inode->i_flags |= S_SYNC;
 451         if (flags & FS_APPEND_FL)
 452                 inode->i_flags |= S_APPEND;
 453         if (flags & FS_IMMUTABLE_FL)
 454                 inode->i_flags |= S_IMMUTABLE;
 455         if (flags & FS_NOATIME_FL)
 456                 inode->i_flags |= S_NOATIME;
 457         if (flags & FS_DIRSYNC_FL)
 458                 inode->i_flags |= S_DIRSYNC;
 459         mapping_set_gfp_mask(inode->i_mapping,
 460                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 461 }
 462
 463 int nilfs_read_inode_common(struct inode *inode,
 464                             struct nilfs_inode *raw_inode)
 465 {
 466         struct nilfs_inode_info *ii = NILFS_I(inode);
 467         int err;
 468
 469         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 470         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 471         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 472         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 473         inode->i_size = le64_to_cpu(raw_inode->i_size);
 474         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 475         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 476         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 477         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 478         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 479         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 480         if (inode->i_nlink == 0)
 481                 return -ESTALE; /* this inode is deleted */
 482
 483         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 484         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 485 #if 0
 486         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 487         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 488                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 489 #endif
 490         ii->i_dir_start_lookup = 0;
 491         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 492
 493         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 494             S_ISLNK(inode->i_mode)) {
 495                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 496                 if (err < 0)
 497                         return err;
 498                 set_bit(NILFS_I_BMAP, &ii->i_state);
 499                 /* No lock is needed; iget() ensures it. */
 500         }
 501         return 0;
 502 }
 503
 504 static int __nilfs_read_inode(struct super_block *sb,
 505                               struct nilfs_root *root, unsigned long ino,
 506                               struct inode *inode)
 507 {
 508         struct the_nilfs *nilfs = sb->s_fs_info;
 509         struct buffer_head *bh;
 510         struct nilfs_inode *raw_inode;
 511         int err;
 512
 513         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 514         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 515         if (unlikely(err))
 516                 goto bad_inode;
 517
 518         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 519
 520         err = nilfs_read_inode_common(inode, raw_inode);
 521         if (err)
 522                 goto failed_unmap;
 523
 524         if (S_ISREG(inode->i_mode)) {
 525                 inode->i_op = &nilfs_file_inode_operations;
 526                 inode->i_fop = &nilfs_file_operations;
 527                 inode->i_mapping->a_ops = &nilfs_aops;
 528         } else if (S_ISDIR(inode->i_mode)) {
 529                 inode->i_op = &nilfs_dir_inode_operations;
 530                 inode->i_fop = &nilfs_dir_operations;
 531                 inode->i_mapping->a_ops = &nilfs_aops;
 532         } else if (S_ISLNK(inode->i_mode)) {
 533                 inode->i_op = &nilfs_symlink_inode_operations;
 534                 inode->i_mapping->a_ops = &nilfs_aops;
 535         } else {
 536                 inode->i_op = &nilfs_special_inode_operations;
 537                 init_special_inode(
 538                         inode, inode->i_mode,
 539                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 540         }
 541         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 542         brelse(bh);
 543         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 544         nilfs_set_inode_flags(inode);
 545         return 0;
 546
 547  failed_unmap:
 548         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 549         brelse(bh);
 550
 551  bad_inode:
 552         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 553         return err;
 554 }
 555
 556 static int nilfs_iget_test(struct inode *inode, void *opaque)
 557 {
 558         struct nilfs_iget_args *args = opaque;
 559         struct nilfs_inode_info *ii;
 560
 561         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 562                 return 0;
 563
 564         ii = NILFS_I(inode);
 565         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 566                 return !args->for_gc;
 567
 568         return args->for_gc && args->cno == ii->i_cno;
 569 }
 570
 571 static int nilfs_iget_set(struct inode *inode, void *opaque)
 572 {
 573         struct nilfs_iget_args *args = opaque;
 574
 575         inode->i_ino = args->ino;
 576         if (args->for_gc) {
 577                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 578                 NILFS_I(inode)->i_cno = args->cno;
 579                 NILFS_I(inode)->i_root = NULL;
 580         } else {
 581                 if (args->root && args->ino == NILFS_ROOT_INO)
 582                         nilfs_get_root(args->root);
 583                 NILFS_I(inode)->i_root = args->root;
 584         }
 585         return 0;
 586 }
 587
 588 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 589                             unsigned long ino)
 590 {
 591         struct nilfs_iget_args args = {
 592                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 593         };
 594
 595         return ilookup5(sb, ino, nilfs_iget_test, &args);
 596 }
 597
 598 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 599                                 unsigned long ino)
 600 {
 601         struct nilfs_iget_args args = {
 602                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 603         };
 604
 605         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 606 }
 607
 608 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 609                          unsigned long ino)
 610 {
 611         struct inode *inode;
 612         int err;
 613
 614         inode = nilfs_iget_locked(sb, root, ino);
 615         if (unlikely(!inode))
 616                 return ERR_PTR(-ENOMEM);
 617         if (!(inode->i_state & I_NEW))
 618                 return inode;
 619
 620         err = __nilfs_read_inode(sb, root, ino, inode);
 621         if (unlikely(err)) {
 622                 iget_failed(inode);
 623                 return ERR_PTR(err);
 624         }
 625         unlock_new_inode(inode);
 626         return inode;
 627 }
 628
 629 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 630                                 __u64 cno)
 631 {
 632         struct nilfs_iget_args args = {
 633                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 634         };
 635         struct inode *inode;
 636         int err;
 637
 638         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 639         if (unlikely(!inode))
 640                 return ERR_PTR(-ENOMEM);
 641         if (!(inode->i_state & I_NEW))
 642                 return inode;
 643
 644         err = nilfs_init_gcinode(inode);
 645         if (unlikely(err)) {
 646                 iget_failed(inode);
 647                 return ERR_PTR(err);
 648         }
 649         unlock_new_inode(inode);
 650         return inode;
 651 }
 652
 653 void nilfs_write_inode_common(struct inode *inode,
 654                               struct nilfs_inode *raw_inode, int has_bmap)
 655 {
 656         struct nilfs_inode_info *ii = NILFS_I(inode);
 657
 658         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 659         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 660         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 661         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 662         raw_inode->i_size = cpu_to_le64(inode->i_size);
 663         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 664         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 665         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 666         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 667         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 668
 669         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 670         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 671
 672         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 673                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 674
 675                 /* zero-fill unused portion in the case of super root block */
 676                 raw_inode->i_xattr = 0;
 677                 raw_inode->i_pad = 0;
 678                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 679                        nilfs->ns_inode_size - sizeof(*raw_inode));
 680         }
 681
 682         if (has_bmap)
 683                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 684         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 685                 raw_inode->i_device_code =
 686                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 687         /* When extending inode, nilfs->ns_inode_size should be checked
 688            for substitutions of appended fields */
 689 }
 690
 691 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
 692 {
 693         ino_t ino = inode->i_ino;
 694         struct nilfs_inode_info *ii = NILFS_I(inode);
 695         struct inode *ifile = ii->i_root->ifile;
 696         struct nilfs_inode *raw_inode;
 697
 698         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 699
 700         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 701                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 702         if (flags & I_DIRTY_DATASYNC)
 703                 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 704
 705         nilfs_write_inode_common(inode, raw_inode, 0);
 706                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 707                    deadlock of bmap. This delays update of i_bmap to just
 708                    before writing */
 709         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 710 }
 711
 712 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 713
 714 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 715                                 unsigned long from)
 716 {
 717         unsigned long b;
 718         int ret;
 719
 720         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 721                 return;
 722 repeat:
 723         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 724         if (ret == -ENOENT)
 725                 return;
 726         else if (ret < 0)
 727                 goto failed;
 728
 729         if (b < from)
 730                 return;
 731
 732         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 733         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 734         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 735         if (!ret || (ret == -ENOMEM &&
 736                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 737                 goto repeat;
 738
 739 failed:
 740         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 741                       "failed to truncate bmap (ino=%lu, err=%d)",
 742                       ii->vfs_inode.i_ino, ret);
 743 }
 744
 745 void nilfs_truncate(struct inode *inode)
 746 {
 747         unsigned long blkoff;
 748         unsigned int blocksize;
 749         struct nilfs_transaction_info ti;
 750         struct super_block *sb = inode->i_sb;
 751         struct nilfs_inode_info *ii = NILFS_I(inode);
 752
 753         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 754                 return;
 755         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 756                 return;
 757
 758         blocksize = sb->s_blocksize;
 759         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 760         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 761
 762         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 763
 764         nilfs_truncate_bmap(ii, blkoff);
 765
 766         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 767         if (IS_SYNC(inode))
 768                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 769
 770         nilfs_mark_inode_dirty(inode);
 771         nilfs_set_file_dirty(inode, 0);
 772         nilfs_transaction_commit(sb);
 773         /* May construct a logical segment and may fail in sync mode.
 774            But truncate has no return value. */
 775 }
 776
 777 static void nilfs_clear_inode(struct inode *inode)
 778 {
 779         struct nilfs_inode_info *ii = NILFS_I(inode);
 780         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 781
 782         /*
 783          * Free resources allocated in nilfs_read_inode(), here.
 784          */
 785         BUG_ON(!list_empty(&ii->i_dirty));
 786         brelse(ii->i_bh);
 787         ii->i_bh = NULL;
 788
 789         if (mdi && mdi->mi_palloc_cache)
 790                 nilfs_palloc_destroy_cache(inode);
 791
 792         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 793                 nilfs_bmap_clear(ii->i_bmap);
 794
 795         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 796
 797         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 798                 nilfs_put_root(ii->i_root);
 799 }
 800
 801 void nilfs_evict_inode(struct inode *inode)
 802 {
 803         struct nilfs_transaction_info ti;
 804         struct super_block *sb = inode->i_sb;
 805         struct nilfs_inode_info *ii = NILFS_I(inode);
 806         int ret;
 807
 808         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 809                 truncate_inode_pages_final(&inode->i_data);
 810                 clear_inode(inode);
 811                 nilfs_clear_inode(inode);
 812                 return;
 813         }
 814         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 815
 816         truncate_inode_pages_final(&inode->i_data);
 817
 818         /* TODO: some of the following operations may fail.  */
 819         nilfs_truncate_bmap(ii, 0);
 820         nilfs_mark_inode_dirty(inode);
 821         clear_inode(inode);
 822
 823         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 824         if (!ret)
 825                 atomic64_dec(&ii->i_root->inodes_count);
 826
 827         nilfs_clear_inode(inode);
 828
 829         if (IS_SYNC(inode))
 830                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 831         nilfs_transaction_commit(sb);
 832         /* May construct a logical segment and may fail in sync mode.
 833            But delete_inode has no return value. */
 834 }
 835
 836 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 837 {
 838         struct nilfs_transaction_info ti;
 839         struct inode *inode = dentry->d_inode;
 840         struct super_block *sb = inode->i_sb;
 841         int err;
 842
 843         err = inode_change_ok(inode, iattr);
 844         if (err)
 845                 return err;
 846
 847         err = nilfs_transaction_begin(sb, &ti, 0);
 848         if (unlikely(err))
 849                 return err;
 850
 851         if ((iattr->ia_valid & ATTR_SIZE) &&
 852             iattr->ia_size != i_size_read(inode)) {
 853                 inode_dio_wait(inode);
 854                 truncate_setsize(inode, iattr->ia_size);
 855                 nilfs_truncate(inode);
 856         }
 857
 858         setattr_copy(inode, iattr);
 859         mark_inode_dirty(inode);
 860
 861         if (iattr->ia_valid & ATTR_MODE) {
 862                 err = nilfs_acl_chmod(inode);
 863                 if (unlikely(err))
 864                         goto out_err;
 865         }
 866
 867         return nilfs_transaction_commit(sb);
 868
 869 out_err:
 870         nilfs_transaction_abort(sb);
 871         return err;
 872 }
 873
 874 int nilfs_permission(struct inode *inode, int mask)
 875 {
 876         struct nilfs_root *root = NILFS_I(inode)->i_root;
 877         if ((mask & MAY_WRITE) && root &&
 878             root->cno != NILFS_CPTREE_CURRENT_CNO)
 879                 return -EROFS; /* snapshot is not writable */
 880
 881         return generic_permission(inode, mask);
 882 }
 883
 884 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 885 {
 886         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 887         struct nilfs_inode_info *ii = NILFS_I(inode);
 888         int err;
 889
 890         spin_lock(&nilfs->ns_inode_lock);
 891         if (ii->i_bh == NULL) {
 892                 spin_unlock(&nilfs->ns_inode_lock);
 893                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 894                                                   inode->i_ino, pbh);
 895                 if (unlikely(err))
 896                         return err;
 897                 spin_lock(&nilfs->ns_inode_lock);
 898                 if (ii->i_bh == NULL)
 899                         ii->i_bh = *pbh;
 900                 else {
 901                         brelse(*pbh);
 902                         *pbh = ii->i_bh;
 903                 }
 904         } else
 905                 *pbh = ii->i_bh;
 906
 907         get_bh(*pbh);
 908         spin_unlock(&nilfs->ns_inode_lock);
 909         return 0;
 910 }
 911
 912 int nilfs_inode_dirty(struct inode *inode)
 913 {
 914         struct nilfs_inode_info *ii = NILFS_I(inode);
 915         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 916         int ret = 0;
 917
 918         if (!list_empty(&ii->i_dirty)) {
 919                 spin_lock(&nilfs->ns_inode_lock);
 920                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 921                         test_bit(NILFS_I_BUSY, &ii->i_state);
 922                 spin_unlock(&nilfs->ns_inode_lock);
 923         }
 924         return ret;
 925 }
 926
 927 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 928 {
 929         struct nilfs_inode_info *ii = NILFS_I(inode);
 930         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 931
 932         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 933
 934         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 935                 return 0;
 936
 937         spin_lock(&nilfs->ns_inode_lock);
 938         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 939             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 940                 /* Because this routine may race with nilfs_dispose_list(),
 941                    we have to check NILFS_I_QUEUED here, too. */
 942                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 943                         /* This will happen when somebody is freeing
 944                            this inode. */
 945                         nilfs_warning(inode->i_sb, __func__,
 946                                       "cannot get inode (ino=%lu)\n",
 947                                       inode->i_ino);
 948                         spin_unlock(&nilfs->ns_inode_lock);
 949                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 950                                            freeing inode */
 951                 }
 952                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 953                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 954         }
 955         spin_unlock(&nilfs->ns_inode_lock);
 956         return 0;
 957 }
 958
 959 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
 960 {
 961         struct buffer_head *ibh;
 962         int err;
 963
 964         err = nilfs_load_inode_block(inode, &ibh);
 965         if (unlikely(err)) {
 966                 nilfs_warning(inode->i_sb, __func__,
 967                               "failed to reget inode block.\n");
 968                 return err;
 969         }
 970         nilfs_update_inode(inode, ibh, flags);
 971         mark_buffer_dirty(ibh);
 972         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 973         brelse(ibh);
 974         return 0;
 975 }
 976
 977 /**
 978  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 979  * @inode: inode of the file to be registered.
 980  *
 981  * nilfs_dirty_inode() loads a inode block containing the specified
 982  * @inode and copies data from a nilfs_inode to a corresponding inode
 983  * entry in the inode block. This operation is excluded from the segment
 984  * construction. This function can be called both as a single operation
 985  * and as a part of indivisible file operations.
 986  */
 987 void nilfs_dirty_inode(struct inode *inode, int flags)
 988 {
 989         struct nilfs_transaction_info ti;
 990         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 991
 992         if (is_bad_inode(inode)) {
 993                 nilfs_warning(inode->i_sb, __func__,
 994                               "tried to mark bad_inode dirty. ignored.\n");
 995                 dump_stack();
 996                 return;
 997         }
 998         if (mdi) {
 999                 nilfs_mdt_mark_dirty(inode);
1000                 return;
1001         }
1002         nilfs_transaction_begin(inode->i_sb, &ti, 0);
1003         __nilfs_mark_inode_dirty(inode, flags);
1004         nilfs_transaction_commit(inode->i_sb); /* never fails */
1005 }
1006
1007 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1008                  __u64 start, __u64 len)
1009 {
1010         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1011         __u64 logical = 0, phys = 0, size = 0;
1012         __u32 flags = 0;
1013         loff_t isize;
1014         sector_t blkoff, end_blkoff;
1015         sector_t delalloc_blkoff;
1016         unsigned long delalloc_blklen;
1017         unsigned int blkbits = inode->i_blkbits;
1018         int ret, n;
1019
1020         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1021         if (ret)
1022                 return ret;
1023
1024         mutex_lock(&inode->i_mutex);
1025
1026         isize = i_size_read(inode);
1027
1028         blkoff = start >> blkbits;
1029         end_blkoff = (start + len - 1) >> blkbits;
1030
1031         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1032                                                         &delalloc_blkoff);
1033
1034         do {
1035                 __u64 blkphy;
1036                 unsigned int maxblocks;
1037
1038                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1039                         if (size) {
1040                                 /* End of the current extent */
1041                                 ret = fiemap_fill_next_extent(
1042                                         fieinfo, logical, phys, size, flags);
1043                                 if (ret)
1044                                         break;
1045                         }
1046                         if (blkoff > end_blkoff)
1047                                 break;
1048
1049                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1050                         logical = blkoff << blkbits;
1051                         phys = 0;
1052                         size = delalloc_blklen << blkbits;
1053
1054                         blkoff = delalloc_blkoff + delalloc_blklen;
1055                         delalloc_blklen = nilfs_find_uncommitted_extent(
1056                                 inode, blkoff, &delalloc_blkoff);
1057                         continue;
1058                 }
1059
1060                 /*
1061                  * Limit the number of blocks that we look up so as
1062                  * not to get into the next delayed allocation extent.
1063                  */
1064                 maxblocks = INT_MAX;
1065                 if (delalloc_blklen)
1066                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1067                                           maxblocks);
1068                 blkphy = 0;
1069
1070                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1071                 n = nilfs_bmap_lookup_contig(
1072                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1073                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1074
1075                 if (n < 0) {
1076                         int past_eof;
1077
1078                         if (unlikely(n != -ENOENT))
1079                                 break; /* error */
1080
1081                         /* HOLE */
1082                         blkoff++;
1083                         past_eof = ((blkoff << blkbits) >= isize);
1084
1085                         if (size) {
1086                                 /* End of the current extent */
1087
1088                                 if (past_eof)
1089                                         flags |= FIEMAP_EXTENT_LAST;
1090
1091                                 ret = fiemap_fill_next_extent(
1092                                         fieinfo, logical, phys, size, flags);
1093                                 if (ret)
1094                                         break;
1095                                 size = 0;
1096                         }
1097                         if (blkoff > end_blkoff || past_eof)
1098                                 break;
1099                 } else {
1100                         if (size) {
1101                                 if (phys && blkphy << blkbits == phys + size) {
1102                                         /* The current extent goes on */
1103                                         size += n << blkbits;
1104                                 } else {
1105                                         /* Terminate the current extent */
1106                                         ret = fiemap_fill_next_extent(
1107                                                 fieinfo, logical, phys, size,
1108                                                 flags);
1109                                         if (ret || blkoff > end_blkoff)
1110                                                 break;
1111
1112                                         /* Start another extent */
1113                                         flags = FIEMAP_EXTENT_MERGED;
1114                                         logical = blkoff << blkbits;
1115                                         phys = blkphy << blkbits;
1116                                         size = n << blkbits;
1117                                 }
1118                         } else {
1119                                 /* Start a new extent */
1120                                 flags = FIEMAP_EXTENT_MERGED;
1121                                 logical = blkoff << blkbits;
1122                                 phys = blkphy << blkbits;
1123                                 size = n << blkbits;
1124                         }
1125                         blkoff += n;
1126                 }
1127                 cond_resched();
1128         } while (true);
1129
1130         /* If ret is 1 then we just hit the end of the extent array */
1131         if (ret == 1)
1132                 ret = 0;
1133
1134         mutex_unlock(&inode->i_mutex);
1135         return ret;
1136 }