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