1 /* 2 * fs/f2fs/file.c 3 * 4 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 5 * http://www.samsung.com/ 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 */ 11 #include <linux/fs.h> 12 #include <linux/f2fs_fs.h> 13 #include <linux/stat.h> 14 #include <linux/buffer_head.h> 15 #include <linux/writeback.h> 16 #include <linux/falloc.h> 17 #include <linux/types.h> 18 #include <linux/uaccess.h> 19 #include <linux/mount.h> 20 21 #include "f2fs.h" 22 #include "node.h" 23 #include "segment.h" 24 #include "xattr.h" 25 #include "acl.h" 26 27 static int f2fs_vm_page_mkwrite(struct vm_area_struct *vma, 28 struct vm_fault *vmf) 29 { 30 struct page *page = vmf->page; 31 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 32 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 33 block_t old_blk_addr; 34 struct dnode_of_data dn; 35 int err; 36 37 f2fs_balance_fs(sbi); 38 39 sb_start_pagefault(inode->i_sb); 40 41 mutex_lock_op(sbi, DATA_NEW); 42 43 /* block allocation */ 44 set_new_dnode(&dn, inode, NULL, NULL, 0); 45 err = get_dnode_of_data(&dn, page->index, 0); 46 if (err) { 47 mutex_unlock_op(sbi, DATA_NEW); 48 goto out; 49 } 50 51 old_blk_addr = dn.data_blkaddr; 52 53 if (old_blk_addr == NULL_ADDR) { 54 err = reserve_new_block(&dn); 55 if (err) { 56 f2fs_put_dnode(&dn); 57 mutex_unlock_op(sbi, DATA_NEW); 58 goto out; 59 } 60 } 61 f2fs_put_dnode(&dn); 62 63 mutex_unlock_op(sbi, DATA_NEW); 64 65 lock_page(page); 66 if (page->mapping != inode->i_mapping || 67 page_offset(page) >= i_size_read(inode) || 68 !PageUptodate(page)) { 69 unlock_page(page); 70 err = -EFAULT; 71 goto out; 72 } 73 74 /* 75 * check to see if the page is mapped already (no holes) 76 */ 77 if (PageMappedToDisk(page)) 78 goto out; 79 80 /* fill the page */ 81 wait_on_page_writeback(page); 82 83 /* page is wholly or partially inside EOF */ 84 if (((page->index + 1) << PAGE_CACHE_SHIFT) > i_size_read(inode)) { 85 unsigned offset; 86 offset = i_size_read(inode) & ~PAGE_CACHE_MASK; 87 zero_user_segment(page, offset, PAGE_CACHE_SIZE); 88 } 89 set_page_dirty(page); 90 SetPageUptodate(page); 91 92 file_update_time(vma->vm_file); 93 out: 94 sb_end_pagefault(inode->i_sb); 95 return block_page_mkwrite_return(err); 96 } 97 98 static const struct vm_operations_struct f2fs_file_vm_ops = { 99 .fault = filemap_fault, 100 .page_mkwrite = f2fs_vm_page_mkwrite, 101 }; 102 103 static int need_to_sync_dir(struct f2fs_sb_info *sbi, struct inode *inode) 104 { 105 struct dentry *dentry; 106 nid_t pino; 107 108 inode = igrab(inode); 109 dentry = d_find_any_alias(inode); 110 if (!dentry) { 111 iput(inode); 112 return 0; 113 } 114 pino = dentry->d_parent->d_inode->i_ino; 115 dput(dentry); 116 iput(inode); 117 return !is_checkpointed_node(sbi, pino); 118 } 119 120 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) 121 { 122 struct inode *inode = file->f_mapping->host; 123 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 124 unsigned long long cur_version; 125 int ret = 0; 126 bool need_cp = false; 127 struct writeback_control wbc = { 128 .sync_mode = WB_SYNC_ALL, 129 .nr_to_write = LONG_MAX, 130 .for_reclaim = 0, 131 }; 132 133 if (inode->i_sb->s_flags & MS_RDONLY) 134 return 0; 135 136 ret = filemap_write_and_wait_range(inode->i_mapping, start, end); 137 if (ret) 138 return ret; 139 140 mutex_lock(&inode->i_mutex); 141 142 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) 143 goto out; 144 145 mutex_lock(&sbi->cp_mutex); 146 cur_version = le64_to_cpu(F2FS_CKPT(sbi)->checkpoint_ver); 147 mutex_unlock(&sbi->cp_mutex); 148 149 if (F2FS_I(inode)->data_version != cur_version && 150 !(inode->i_state & I_DIRTY)) 151 goto out; 152 F2FS_I(inode)->data_version--; 153 154 if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1) 155 need_cp = true; 156 if (is_inode_flag_set(F2FS_I(inode), FI_NEED_CP)) 157 need_cp = true; 158 if (!space_for_roll_forward(sbi)) 159 need_cp = true; 160 if (need_to_sync_dir(sbi, inode)) 161 need_cp = true; 162 163 f2fs_write_inode(inode, NULL); 164 165 if (need_cp) { 166 /* all the dirty node pages should be flushed for POR */ 167 ret = f2fs_sync_fs(inode->i_sb, 1); 168 clear_inode_flag(F2FS_I(inode), FI_NEED_CP); 169 } else { 170 while (sync_node_pages(sbi, inode->i_ino, &wbc) == 0) 171 f2fs_write_inode(inode, NULL); 172 filemap_fdatawait_range(sbi->node_inode->i_mapping, 173 0, LONG_MAX); 174 } 175 out: 176 mutex_unlock(&inode->i_mutex); 177 return ret; 178 } 179 180 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma) 181 { 182 file_accessed(file); 183 vma->vm_ops = &f2fs_file_vm_ops; 184 return 0; 185 } 186 187 static int truncate_data_blocks_range(struct dnode_of_data *dn, int count) 188 { 189 int nr_free = 0, ofs = dn->ofs_in_node; 190 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 191 struct f2fs_node *raw_node; 192 __le32 *addr; 193 194 raw_node = page_address(dn->node_page); 195 addr = blkaddr_in_node(raw_node) + ofs; 196 197 for ( ; count > 0; count--, addr++, dn->ofs_in_node++) { 198 block_t blkaddr = le32_to_cpu(*addr); 199 if (blkaddr == NULL_ADDR) 200 continue; 201 202 update_extent_cache(NULL_ADDR, dn); 203 invalidate_blocks(sbi, blkaddr); 204 dec_valid_block_count(sbi, dn->inode, 1); 205 nr_free++; 206 } 207 if (nr_free) { 208 set_page_dirty(dn->node_page); 209 sync_inode_page(dn); 210 } 211 dn->ofs_in_node = ofs; 212 return nr_free; 213 } 214 215 void truncate_data_blocks(struct dnode_of_data *dn) 216 { 217 truncate_data_blocks_range(dn, ADDRS_PER_BLOCK); 218 } 219 220 static void truncate_partial_data_page(struct inode *inode, u64 from) 221 { 222 unsigned offset = from & (PAGE_CACHE_SIZE - 1); 223 struct page *page; 224 225 if (!offset) 226 return; 227 228 page = find_data_page(inode, from >> PAGE_CACHE_SHIFT); 229 if (IS_ERR(page)) 230 return; 231 232 lock_page(page); 233 wait_on_page_writeback(page); 234 zero_user(page, offset, PAGE_CACHE_SIZE - offset); 235 set_page_dirty(page); 236 f2fs_put_page(page, 1); 237 } 238 239 static int truncate_blocks(struct inode *inode, u64 from) 240 { 241 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 242 unsigned int blocksize = inode->i_sb->s_blocksize; 243 struct dnode_of_data dn; 244 pgoff_t free_from; 245 int count = 0; 246 int err; 247 248 free_from = (pgoff_t) 249 ((from + blocksize - 1) >> (sbi->log_blocksize)); 250 251 mutex_lock_op(sbi, DATA_TRUNC); 252 253 set_new_dnode(&dn, inode, NULL, NULL, 0); 254 err = get_dnode_of_data(&dn, free_from, RDONLY_NODE); 255 if (err) { 256 if (err == -ENOENT) 257 goto free_next; 258 mutex_unlock_op(sbi, DATA_TRUNC); 259 return err; 260 } 261 262 if (IS_INODE(dn.node_page)) 263 count = ADDRS_PER_INODE; 264 else 265 count = ADDRS_PER_BLOCK; 266 267 count -= dn.ofs_in_node; 268 BUG_ON(count < 0); 269 if (dn.ofs_in_node || IS_INODE(dn.node_page)) { 270 truncate_data_blocks_range(&dn, count); 271 free_from += count; 272 } 273 274 f2fs_put_dnode(&dn); 275 free_next: 276 err = truncate_inode_blocks(inode, free_from); 277 mutex_unlock_op(sbi, DATA_TRUNC); 278 279 /* lastly zero out the first data page */ 280 truncate_partial_data_page(inode, from); 281 282 return err; 283 } 284 285 void f2fs_truncate(struct inode *inode) 286 { 287 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 288 S_ISLNK(inode->i_mode))) 289 return; 290 291 if (!truncate_blocks(inode, i_size_read(inode))) { 292 inode->i_mtime = inode->i_ctime = CURRENT_TIME; 293 mark_inode_dirty(inode); 294 } 295 296 f2fs_balance_fs(F2FS_SB(inode->i_sb)); 297 } 298 299 static int f2fs_getattr(struct vfsmount *mnt, 300 struct dentry *dentry, struct kstat *stat) 301 { 302 struct inode *inode = dentry->d_inode; 303 generic_fillattr(inode, stat); 304 stat->blocks <<= 3; 305 return 0; 306 } 307 308 #ifdef CONFIG_F2FS_FS_POSIX_ACL 309 static void __setattr_copy(struct inode *inode, const struct iattr *attr) 310 { 311 struct f2fs_inode_info *fi = F2FS_I(inode); 312 unsigned int ia_valid = attr->ia_valid; 313 314 if (ia_valid & ATTR_UID) 315 inode->i_uid = attr->ia_uid; 316 if (ia_valid & ATTR_GID) 317 inode->i_gid = attr->ia_gid; 318 if (ia_valid & ATTR_ATIME) 319 inode->i_atime = timespec_trunc(attr->ia_atime, 320 inode->i_sb->s_time_gran); 321 if (ia_valid & ATTR_MTIME) 322 inode->i_mtime = timespec_trunc(attr->ia_mtime, 323 inode->i_sb->s_time_gran); 324 if (ia_valid & ATTR_CTIME) 325 inode->i_ctime = timespec_trunc(attr->ia_ctime, 326 inode->i_sb->s_time_gran); 327 if (ia_valid & ATTR_MODE) { 328 umode_t mode = attr->ia_mode; 329 330 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) 331 mode &= ~S_ISGID; 332 set_acl_inode(fi, mode); 333 } 334 } 335 #else 336 #define __setattr_copy setattr_copy 337 #endif 338 339 int f2fs_setattr(struct dentry *dentry, struct iattr *attr) 340 { 341 struct inode *inode = dentry->d_inode; 342 struct f2fs_inode_info *fi = F2FS_I(inode); 343 int err; 344 345 err = inode_change_ok(inode, attr); 346 if (err) 347 return err; 348 349 if ((attr->ia_valid & ATTR_SIZE) && 350 attr->ia_size != i_size_read(inode)) { 351 truncate_setsize(inode, attr->ia_size); 352 f2fs_truncate(inode); 353 } 354 355 __setattr_copy(inode, attr); 356 357 if (attr->ia_valid & ATTR_MODE) { 358 err = f2fs_acl_chmod(inode); 359 if (err || is_inode_flag_set(fi, FI_ACL_MODE)) { 360 inode->i_mode = fi->i_acl_mode; 361 clear_inode_flag(fi, FI_ACL_MODE); 362 } 363 } 364 365 mark_inode_dirty(inode); 366 return err; 367 } 368 369 const struct inode_operations f2fs_file_inode_operations = { 370 .getattr = f2fs_getattr, 371 .setattr = f2fs_setattr, 372 .get_acl = f2fs_get_acl, 373 #ifdef CONFIG_F2FS_FS_XATTR 374 .setxattr = generic_setxattr, 375 .getxattr = generic_getxattr, 376 .listxattr = f2fs_listxattr, 377 .removexattr = generic_removexattr, 378 #endif 379 }; 380 381 static void fill_zero(struct inode *inode, pgoff_t index, 382 loff_t start, loff_t len) 383 { 384 struct page *page; 385 386 if (!len) 387 return; 388 389 page = get_new_data_page(inode, index, false); 390 391 if (!IS_ERR(page)) { 392 wait_on_page_writeback(page); 393 zero_user(page, start, len); 394 set_page_dirty(page); 395 f2fs_put_page(page, 1); 396 } 397 } 398 399 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end) 400 { 401 pgoff_t index; 402 int err; 403 404 for (index = pg_start; index < pg_end; index++) { 405 struct dnode_of_data dn; 406 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 407 408 mutex_lock_op(sbi, DATA_TRUNC); 409 set_new_dnode(&dn, inode, NULL, NULL, 0); 410 err = get_dnode_of_data(&dn, index, RDONLY_NODE); 411 if (err) { 412 mutex_unlock_op(sbi, DATA_TRUNC); 413 if (err == -ENOENT) 414 continue; 415 return err; 416 } 417 418 if (dn.data_blkaddr != NULL_ADDR) 419 truncate_data_blocks_range(&dn, 1); 420 f2fs_put_dnode(&dn); 421 mutex_unlock_op(sbi, DATA_TRUNC); 422 } 423 return 0; 424 } 425 426 static int punch_hole(struct inode *inode, loff_t offset, loff_t len, int mode) 427 { 428 pgoff_t pg_start, pg_end; 429 loff_t off_start, off_end; 430 int ret = 0; 431 432 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT; 433 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT; 434 435 off_start = offset & (PAGE_CACHE_SIZE - 1); 436 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1); 437 438 if (pg_start == pg_end) { 439 fill_zero(inode, pg_start, off_start, 440 off_end - off_start); 441 } else { 442 if (off_start) 443 fill_zero(inode, pg_start++, off_start, 444 PAGE_CACHE_SIZE - off_start); 445 if (off_end) 446 fill_zero(inode, pg_end, 0, off_end); 447 448 if (pg_start < pg_end) { 449 struct address_space *mapping = inode->i_mapping; 450 loff_t blk_start, blk_end; 451 452 blk_start = pg_start << PAGE_CACHE_SHIFT; 453 blk_end = pg_end << PAGE_CACHE_SHIFT; 454 truncate_inode_pages_range(mapping, blk_start, 455 blk_end - 1); 456 ret = truncate_hole(inode, pg_start, pg_end); 457 } 458 } 459 460 if (!(mode & FALLOC_FL_KEEP_SIZE) && 461 i_size_read(inode) <= (offset + len)) { 462 i_size_write(inode, offset); 463 mark_inode_dirty(inode); 464 } 465 466 return ret; 467 } 468 469 static int expand_inode_data(struct inode *inode, loff_t offset, 470 loff_t len, int mode) 471 { 472 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 473 pgoff_t index, pg_start, pg_end; 474 loff_t new_size = i_size_read(inode); 475 loff_t off_start, off_end; 476 int ret = 0; 477 478 ret = inode_newsize_ok(inode, (len + offset)); 479 if (ret) 480 return ret; 481 482 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT; 483 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT; 484 485 off_start = offset & (PAGE_CACHE_SIZE - 1); 486 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1); 487 488 for (index = pg_start; index <= pg_end; index++) { 489 struct dnode_of_data dn; 490 491 mutex_lock_op(sbi, DATA_NEW); 492 493 set_new_dnode(&dn, inode, NULL, NULL, 0); 494 ret = get_dnode_of_data(&dn, index, 0); 495 if (ret) { 496 mutex_unlock_op(sbi, DATA_NEW); 497 break; 498 } 499 500 if (dn.data_blkaddr == NULL_ADDR) { 501 ret = reserve_new_block(&dn); 502 if (ret) { 503 f2fs_put_dnode(&dn); 504 mutex_unlock_op(sbi, DATA_NEW); 505 break; 506 } 507 } 508 f2fs_put_dnode(&dn); 509 510 mutex_unlock_op(sbi, DATA_NEW); 511 512 if (pg_start == pg_end) 513 new_size = offset + len; 514 else if (index == pg_start && off_start) 515 new_size = (index + 1) << PAGE_CACHE_SHIFT; 516 else if (index == pg_end) 517 new_size = (index << PAGE_CACHE_SHIFT) + off_end; 518 else 519 new_size += PAGE_CACHE_SIZE; 520 } 521 522 if (!(mode & FALLOC_FL_KEEP_SIZE) && 523 i_size_read(inode) < new_size) { 524 i_size_write(inode, new_size); 525 mark_inode_dirty(inode); 526 } 527 528 return ret; 529 } 530 531 static long f2fs_fallocate(struct file *file, int mode, 532 loff_t offset, loff_t len) 533 { 534 struct inode *inode = file->f_path.dentry->d_inode; 535 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 536 long ret; 537 538 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 539 return -EOPNOTSUPP; 540 541 if (mode & FALLOC_FL_PUNCH_HOLE) 542 ret = punch_hole(inode, offset, len, mode); 543 else 544 ret = expand_inode_data(inode, offset, len, mode); 545 546 f2fs_balance_fs(sbi); 547 return ret; 548 } 549 550 #define F2FS_REG_FLMASK (~(FS_DIRSYNC_FL | FS_TOPDIR_FL)) 551 #define F2FS_OTHER_FLMASK (FS_NODUMP_FL | FS_NOATIME_FL) 552 553 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags) 554 { 555 if (S_ISDIR(mode)) 556 return flags; 557 else if (S_ISREG(mode)) 558 return flags & F2FS_REG_FLMASK; 559 else 560 return flags & F2FS_OTHER_FLMASK; 561 } 562 563 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 564 { 565 struct inode *inode = filp->f_dentry->d_inode; 566 struct f2fs_inode_info *fi = F2FS_I(inode); 567 unsigned int flags; 568 int ret; 569 570 switch (cmd) { 571 case FS_IOC_GETFLAGS: 572 flags = fi->i_flags & FS_FL_USER_VISIBLE; 573 return put_user(flags, (int __user *) arg); 574 case FS_IOC_SETFLAGS: 575 { 576 unsigned int oldflags; 577 578 ret = mnt_want_write(filp->f_path.mnt); 579 if (ret) 580 return ret; 581 582 if (!inode_owner_or_capable(inode)) { 583 ret = -EACCES; 584 goto out; 585 } 586 587 if (get_user(flags, (int __user *) arg)) { 588 ret = -EFAULT; 589 goto out; 590 } 591 592 flags = f2fs_mask_flags(inode->i_mode, flags); 593 594 mutex_lock(&inode->i_mutex); 595 596 oldflags = fi->i_flags; 597 598 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 599 if (!capable(CAP_LINUX_IMMUTABLE)) { 600 mutex_unlock(&inode->i_mutex); 601 ret = -EPERM; 602 goto out; 603 } 604 } 605 606 flags = flags & FS_FL_USER_MODIFIABLE; 607 flags |= oldflags & ~FS_FL_USER_MODIFIABLE; 608 fi->i_flags = flags; 609 mutex_unlock(&inode->i_mutex); 610 611 f2fs_set_inode_flags(inode); 612 inode->i_ctime = CURRENT_TIME; 613 mark_inode_dirty(inode); 614 out: 615 mnt_drop_write(filp->f_path.mnt); 616 return ret; 617 } 618 default: 619 return -ENOTTY; 620 } 621 } 622 623 const struct file_operations f2fs_file_operations = { 624 .llseek = generic_file_llseek, 625 .read = do_sync_read, 626 .write = do_sync_write, 627 .aio_read = generic_file_aio_read, 628 .aio_write = generic_file_aio_write, 629 .open = generic_file_open, 630 .mmap = f2fs_file_mmap, 631 .fsync = f2fs_sync_file, 632 .fallocate = f2fs_fallocate, 633 .unlocked_ioctl = f2fs_ioctl, 634 .splice_read = generic_file_splice_read, 635 .splice_write = generic_file_splice_write, 636 }; 637