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