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/blkdev.h> 17 #include <linux/falloc.h> 18 #include <linux/types.h> 19 #include <linux/compat.h> 20 #include <linux/uaccess.h> 21 #include <linux/mount.h> 22 23 #include "f2fs.h" 24 #include "node.h" 25 #include "segment.h" 26 #include "xattr.h" 27 #include "acl.h" 28 #include <trace/events/f2fs.h> 29 30 static int f2fs_vm_page_mkwrite(struct vm_area_struct *vma, 31 struct vm_fault *vmf) 32 { 33 struct page *page = vmf->page; 34 struct inode *inode = file_inode(vma->vm_file); 35 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 36 block_t old_blk_addr; 37 struct dnode_of_data dn; 38 int err; 39 40 f2fs_balance_fs(sbi); 41 42 sb_start_pagefault(inode->i_sb); 43 44 /* block allocation */ 45 f2fs_lock_op(sbi); 46 set_new_dnode(&dn, inode, NULL, NULL, 0); 47 err = get_dnode_of_data(&dn, page->index, ALLOC_NODE); 48 if (err) { 49 f2fs_unlock_op(sbi); 50 goto out; 51 } 52 53 old_blk_addr = dn.data_blkaddr; 54 55 if (old_blk_addr == NULL_ADDR) { 56 err = reserve_new_block(&dn); 57 if (err) { 58 f2fs_put_dnode(&dn); 59 f2fs_unlock_op(sbi); 60 goto out; 61 } 62 } 63 f2fs_put_dnode(&dn); 64 f2fs_unlock_op(sbi); 65 66 file_update_time(vma->vm_file); 67 lock_page(page); 68 if (page->mapping != inode->i_mapping || 69 page_offset(page) > i_size_read(inode) || 70 !PageUptodate(page)) { 71 unlock_page(page); 72 err = -EFAULT; 73 goto out; 74 } 75 76 /* 77 * check to see if the page is mapped already (no holes) 78 */ 79 if (PageMappedToDisk(page)) 80 goto mapped; 81 82 /* page is wholly or partially inside EOF */ 83 if (((page->index + 1) << PAGE_CACHE_SHIFT) > i_size_read(inode)) { 84 unsigned offset; 85 offset = i_size_read(inode) & ~PAGE_CACHE_MASK; 86 zero_user_segment(page, offset, PAGE_CACHE_SIZE); 87 } 88 set_page_dirty(page); 89 SetPageUptodate(page); 90 91 trace_f2fs_vm_page_mkwrite(page, DATA); 92 mapped: 93 /* fill the page */ 94 wait_on_page_writeback(page); 95 out: 96 sb_end_pagefault(inode->i_sb); 97 return block_page_mkwrite_return(err); 98 } 99 100 static const struct vm_operations_struct f2fs_file_vm_ops = { 101 .fault = filemap_fault, 102 .page_mkwrite = f2fs_vm_page_mkwrite, 103 .remap_pages = generic_file_remap_pages, 104 }; 105 106 static int get_parent_ino(struct inode *inode, nid_t *pino) 107 { 108 struct dentry *dentry; 109 110 inode = igrab(inode); 111 dentry = d_find_any_alias(inode); 112 iput(inode); 113 if (!dentry) 114 return 0; 115 116 if (update_dent_inode(inode, &dentry->d_name)) { 117 dput(dentry); 118 return 0; 119 } 120 121 *pino = parent_ino(dentry); 122 dput(dentry); 123 return 1; 124 } 125 126 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) 127 { 128 struct inode *inode = file->f_mapping->host; 129 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 130 int ret = 0; 131 bool need_cp = false; 132 struct writeback_control wbc = { 133 .sync_mode = WB_SYNC_ALL, 134 .nr_to_write = LONG_MAX, 135 .for_reclaim = 0, 136 }; 137 138 if (f2fs_readonly(inode->i_sb)) 139 return 0; 140 141 trace_f2fs_sync_file_enter(inode); 142 ret = filemap_write_and_wait_range(inode->i_mapping, start, end); 143 if (ret) { 144 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret); 145 return ret; 146 } 147 148 /* guarantee free sections for fsync */ 149 f2fs_balance_fs(sbi); 150 151 mutex_lock(&inode->i_mutex); 152 153 /* 154 * Both of fdatasync() and fsync() are able to be recovered from 155 * sudden-power-off. 156 */ 157 if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1) 158 need_cp = true; 159 else if (file_wrong_pino(inode)) 160 need_cp = true; 161 else if (!space_for_roll_forward(sbi)) 162 need_cp = true; 163 else if (!is_checkpointed_node(sbi, F2FS_I(inode)->i_pino)) 164 need_cp = true; 165 else if (F2FS_I(inode)->xattr_ver == cur_cp_version(F2FS_CKPT(sbi))) 166 need_cp = true; 167 168 if (need_cp) { 169 nid_t pino; 170 171 F2FS_I(inode)->xattr_ver = 0; 172 173 /* all the dirty node pages should be flushed for POR */ 174 ret = f2fs_sync_fs(inode->i_sb, 1); 175 if (file_wrong_pino(inode) && inode->i_nlink == 1 && 176 get_parent_ino(inode, &pino)) { 177 F2FS_I(inode)->i_pino = pino; 178 file_got_pino(inode); 179 mark_inode_dirty_sync(inode); 180 ret = f2fs_write_inode(inode, NULL); 181 if (ret) 182 goto out; 183 } 184 } else { 185 /* if there is no written node page, write its inode page */ 186 while (!sync_node_pages(sbi, inode->i_ino, &wbc)) { 187 mark_inode_dirty_sync(inode); 188 ret = f2fs_write_inode(inode, NULL); 189 if (ret) 190 goto out; 191 } 192 ret = wait_on_node_pages_writeback(sbi, inode->i_ino); 193 if (ret) 194 goto out; 195 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); 196 } 197 out: 198 mutex_unlock(&inode->i_mutex); 199 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret); 200 return ret; 201 } 202 203 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma) 204 { 205 file_accessed(file); 206 vma->vm_ops = &f2fs_file_vm_ops; 207 return 0; 208 } 209 210 int truncate_data_blocks_range(struct dnode_of_data *dn, int count) 211 { 212 int nr_free = 0, ofs = dn->ofs_in_node; 213 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 214 struct f2fs_node *raw_node; 215 __le32 *addr; 216 217 raw_node = F2FS_NODE(dn->node_page); 218 addr = blkaddr_in_node(raw_node) + ofs; 219 220 for ( ; count > 0; count--, addr++, dn->ofs_in_node++) { 221 block_t blkaddr = le32_to_cpu(*addr); 222 if (blkaddr == NULL_ADDR) 223 continue; 224 225 update_extent_cache(NULL_ADDR, dn); 226 invalidate_blocks(sbi, blkaddr); 227 nr_free++; 228 } 229 if (nr_free) { 230 dec_valid_block_count(sbi, dn->inode, nr_free); 231 set_page_dirty(dn->node_page); 232 sync_inode_page(dn); 233 } 234 dn->ofs_in_node = ofs; 235 236 trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid, 237 dn->ofs_in_node, nr_free); 238 return nr_free; 239 } 240 241 void truncate_data_blocks(struct dnode_of_data *dn) 242 { 243 truncate_data_blocks_range(dn, ADDRS_PER_BLOCK); 244 } 245 246 static void truncate_partial_data_page(struct inode *inode, u64 from) 247 { 248 unsigned offset = from & (PAGE_CACHE_SIZE - 1); 249 struct page *page; 250 251 if (!offset) 252 return; 253 254 page = find_data_page(inode, from >> PAGE_CACHE_SHIFT, false); 255 if (IS_ERR(page)) 256 return; 257 258 lock_page(page); 259 if (page->mapping != inode->i_mapping) { 260 f2fs_put_page(page, 1); 261 return; 262 } 263 wait_on_page_writeback(page); 264 zero_user(page, offset, PAGE_CACHE_SIZE - offset); 265 set_page_dirty(page); 266 f2fs_put_page(page, 1); 267 } 268 269 static int truncate_blocks(struct inode *inode, u64 from) 270 { 271 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 272 unsigned int blocksize = inode->i_sb->s_blocksize; 273 struct dnode_of_data dn; 274 pgoff_t free_from; 275 int count = 0; 276 int err; 277 278 trace_f2fs_truncate_blocks_enter(inode, from); 279 280 free_from = (pgoff_t) 281 ((from + blocksize - 1) >> (sbi->log_blocksize)); 282 283 f2fs_lock_op(sbi); 284 set_new_dnode(&dn, inode, NULL, NULL, 0); 285 err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE); 286 if (err) { 287 if (err == -ENOENT) 288 goto free_next; 289 f2fs_unlock_op(sbi); 290 trace_f2fs_truncate_blocks_exit(inode, err); 291 return err; 292 } 293 294 if (IS_INODE(dn.node_page)) 295 count = ADDRS_PER_INODE(F2FS_I(inode)); 296 else 297 count = ADDRS_PER_BLOCK; 298 299 count -= dn.ofs_in_node; 300 f2fs_bug_on(count < 0); 301 302 if (dn.ofs_in_node || IS_INODE(dn.node_page)) { 303 truncate_data_blocks_range(&dn, count); 304 free_from += count; 305 } 306 307 f2fs_put_dnode(&dn); 308 free_next: 309 err = truncate_inode_blocks(inode, free_from); 310 f2fs_unlock_op(sbi); 311 312 /* lastly zero out the first data page */ 313 truncate_partial_data_page(inode, from); 314 315 trace_f2fs_truncate_blocks_exit(inode, err); 316 return err; 317 } 318 319 void f2fs_truncate(struct inode *inode) 320 { 321 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 322 S_ISLNK(inode->i_mode))) 323 return; 324 325 trace_f2fs_truncate(inode); 326 327 if (!truncate_blocks(inode, i_size_read(inode))) { 328 inode->i_mtime = inode->i_ctime = CURRENT_TIME; 329 mark_inode_dirty(inode); 330 } 331 } 332 333 int f2fs_getattr(struct vfsmount *mnt, 334 struct dentry *dentry, struct kstat *stat) 335 { 336 struct inode *inode = dentry->d_inode; 337 generic_fillattr(inode, stat); 338 stat->blocks <<= 3; 339 return 0; 340 } 341 342 #ifdef CONFIG_F2FS_FS_POSIX_ACL 343 static void __setattr_copy(struct inode *inode, const struct iattr *attr) 344 { 345 struct f2fs_inode_info *fi = F2FS_I(inode); 346 unsigned int ia_valid = attr->ia_valid; 347 348 if (ia_valid & ATTR_UID) 349 inode->i_uid = attr->ia_uid; 350 if (ia_valid & ATTR_GID) 351 inode->i_gid = attr->ia_gid; 352 if (ia_valid & ATTR_ATIME) 353 inode->i_atime = timespec_trunc(attr->ia_atime, 354 inode->i_sb->s_time_gran); 355 if (ia_valid & ATTR_MTIME) 356 inode->i_mtime = timespec_trunc(attr->ia_mtime, 357 inode->i_sb->s_time_gran); 358 if (ia_valid & ATTR_CTIME) 359 inode->i_ctime = timespec_trunc(attr->ia_ctime, 360 inode->i_sb->s_time_gran); 361 if (ia_valid & ATTR_MODE) { 362 umode_t mode = attr->ia_mode; 363 364 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) 365 mode &= ~S_ISGID; 366 set_acl_inode(fi, mode); 367 } 368 } 369 #else 370 #define __setattr_copy setattr_copy 371 #endif 372 373 int f2fs_setattr(struct dentry *dentry, struct iattr *attr) 374 { 375 struct inode *inode = dentry->d_inode; 376 struct f2fs_inode_info *fi = F2FS_I(inode); 377 int err; 378 379 err = inode_change_ok(inode, attr); 380 if (err) 381 return err; 382 383 if ((attr->ia_valid & ATTR_SIZE) && 384 attr->ia_size != i_size_read(inode)) { 385 truncate_setsize(inode, attr->ia_size); 386 f2fs_truncate(inode); 387 f2fs_balance_fs(F2FS_SB(inode->i_sb)); 388 } 389 390 __setattr_copy(inode, attr); 391 392 if (attr->ia_valid & ATTR_MODE) { 393 err = f2fs_acl_chmod(inode); 394 if (err || is_inode_flag_set(fi, FI_ACL_MODE)) { 395 inode->i_mode = fi->i_acl_mode; 396 clear_inode_flag(fi, FI_ACL_MODE); 397 } 398 } 399 400 mark_inode_dirty(inode); 401 return err; 402 } 403 404 const struct inode_operations f2fs_file_inode_operations = { 405 .getattr = f2fs_getattr, 406 .setattr = f2fs_setattr, 407 .get_acl = f2fs_get_acl, 408 #ifdef CONFIG_F2FS_FS_XATTR 409 .setxattr = generic_setxattr, 410 .getxattr = generic_getxattr, 411 .listxattr = f2fs_listxattr, 412 .removexattr = generic_removexattr, 413 #endif 414 }; 415 416 static void fill_zero(struct inode *inode, pgoff_t index, 417 loff_t start, loff_t len) 418 { 419 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 420 struct page *page; 421 422 if (!len) 423 return; 424 425 f2fs_balance_fs(sbi); 426 427 f2fs_lock_op(sbi); 428 page = get_new_data_page(inode, NULL, index, false); 429 f2fs_unlock_op(sbi); 430 431 if (!IS_ERR(page)) { 432 wait_on_page_writeback(page); 433 zero_user(page, start, len); 434 set_page_dirty(page); 435 f2fs_put_page(page, 1); 436 } 437 } 438 439 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end) 440 { 441 pgoff_t index; 442 int err; 443 444 for (index = pg_start; index < pg_end; index++) { 445 struct dnode_of_data dn; 446 447 set_new_dnode(&dn, inode, NULL, NULL, 0); 448 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 449 if (err) { 450 if (err == -ENOENT) 451 continue; 452 return err; 453 } 454 455 if (dn.data_blkaddr != NULL_ADDR) 456 truncate_data_blocks_range(&dn, 1); 457 f2fs_put_dnode(&dn); 458 } 459 return 0; 460 } 461 462 static int punch_hole(struct inode *inode, loff_t offset, loff_t len, int mode) 463 { 464 pgoff_t pg_start, pg_end; 465 loff_t off_start, off_end; 466 int ret = 0; 467 468 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT; 469 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT; 470 471 off_start = offset & (PAGE_CACHE_SIZE - 1); 472 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1); 473 474 if (pg_start == pg_end) { 475 fill_zero(inode, pg_start, off_start, 476 off_end - off_start); 477 } else { 478 if (off_start) 479 fill_zero(inode, pg_start++, off_start, 480 PAGE_CACHE_SIZE - off_start); 481 if (off_end) 482 fill_zero(inode, pg_end, 0, off_end); 483 484 if (pg_start < pg_end) { 485 struct address_space *mapping = inode->i_mapping; 486 loff_t blk_start, blk_end; 487 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 488 489 f2fs_balance_fs(sbi); 490 491 blk_start = pg_start << PAGE_CACHE_SHIFT; 492 blk_end = pg_end << PAGE_CACHE_SHIFT; 493 truncate_inode_pages_range(mapping, blk_start, 494 blk_end - 1); 495 496 f2fs_lock_op(sbi); 497 ret = truncate_hole(inode, pg_start, pg_end); 498 f2fs_unlock_op(sbi); 499 } 500 } 501 502 if (!(mode & FALLOC_FL_KEEP_SIZE) && 503 i_size_read(inode) <= (offset + len)) { 504 i_size_write(inode, offset); 505 mark_inode_dirty(inode); 506 } 507 508 return ret; 509 } 510 511 static int expand_inode_data(struct inode *inode, loff_t offset, 512 loff_t len, int mode) 513 { 514 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 515 pgoff_t index, pg_start, pg_end; 516 loff_t new_size = i_size_read(inode); 517 loff_t off_start, off_end; 518 int ret = 0; 519 520 ret = inode_newsize_ok(inode, (len + offset)); 521 if (ret) 522 return ret; 523 524 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT; 525 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT; 526 527 off_start = offset & (PAGE_CACHE_SIZE - 1); 528 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1); 529 530 for (index = pg_start; index <= pg_end; index++) { 531 struct dnode_of_data dn; 532 533 f2fs_lock_op(sbi); 534 set_new_dnode(&dn, inode, NULL, NULL, 0); 535 ret = get_dnode_of_data(&dn, index, ALLOC_NODE); 536 if (ret) { 537 f2fs_unlock_op(sbi); 538 break; 539 } 540 541 if (dn.data_blkaddr == NULL_ADDR) { 542 ret = reserve_new_block(&dn); 543 if (ret) { 544 f2fs_put_dnode(&dn); 545 f2fs_unlock_op(sbi); 546 break; 547 } 548 } 549 f2fs_put_dnode(&dn); 550 f2fs_unlock_op(sbi); 551 552 if (pg_start == pg_end) 553 new_size = offset + len; 554 else if (index == pg_start && off_start) 555 new_size = (index + 1) << PAGE_CACHE_SHIFT; 556 else if (index == pg_end) 557 new_size = (index << PAGE_CACHE_SHIFT) + off_end; 558 else 559 new_size += PAGE_CACHE_SIZE; 560 } 561 562 if (!(mode & FALLOC_FL_KEEP_SIZE) && 563 i_size_read(inode) < new_size) { 564 i_size_write(inode, new_size); 565 mark_inode_dirty(inode); 566 } 567 568 return ret; 569 } 570 571 static long f2fs_fallocate(struct file *file, int mode, 572 loff_t offset, loff_t len) 573 { 574 struct inode *inode = file_inode(file); 575 long ret; 576 577 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 578 return -EOPNOTSUPP; 579 580 if (mode & FALLOC_FL_PUNCH_HOLE) 581 ret = punch_hole(inode, offset, len, mode); 582 else 583 ret = expand_inode_data(inode, offset, len, mode); 584 585 if (!ret) { 586 inode->i_mtime = inode->i_ctime = CURRENT_TIME; 587 mark_inode_dirty(inode); 588 } 589 trace_f2fs_fallocate(inode, mode, offset, len, ret); 590 return ret; 591 } 592 593 #define F2FS_REG_FLMASK (~(FS_DIRSYNC_FL | FS_TOPDIR_FL)) 594 #define F2FS_OTHER_FLMASK (FS_NODUMP_FL | FS_NOATIME_FL) 595 596 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags) 597 { 598 if (S_ISDIR(mode)) 599 return flags; 600 else if (S_ISREG(mode)) 601 return flags & F2FS_REG_FLMASK; 602 else 603 return flags & F2FS_OTHER_FLMASK; 604 } 605 606 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 607 { 608 struct inode *inode = file_inode(filp); 609 struct f2fs_inode_info *fi = F2FS_I(inode); 610 unsigned int flags; 611 int ret; 612 613 switch (cmd) { 614 case F2FS_IOC_GETFLAGS: 615 flags = fi->i_flags & FS_FL_USER_VISIBLE; 616 return put_user(flags, (int __user *) arg); 617 case F2FS_IOC_SETFLAGS: 618 { 619 unsigned int oldflags; 620 621 ret = mnt_want_write_file(filp); 622 if (ret) 623 return ret; 624 625 if (!inode_owner_or_capable(inode)) { 626 ret = -EACCES; 627 goto out; 628 } 629 630 if (get_user(flags, (int __user *) arg)) { 631 ret = -EFAULT; 632 goto out; 633 } 634 635 flags = f2fs_mask_flags(inode->i_mode, flags); 636 637 mutex_lock(&inode->i_mutex); 638 639 oldflags = fi->i_flags; 640 641 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 642 if (!capable(CAP_LINUX_IMMUTABLE)) { 643 mutex_unlock(&inode->i_mutex); 644 ret = -EPERM; 645 goto out; 646 } 647 } 648 649 flags = flags & FS_FL_USER_MODIFIABLE; 650 flags |= oldflags & ~FS_FL_USER_MODIFIABLE; 651 fi->i_flags = flags; 652 mutex_unlock(&inode->i_mutex); 653 654 f2fs_set_inode_flags(inode); 655 inode->i_ctime = CURRENT_TIME; 656 mark_inode_dirty(inode); 657 out: 658 mnt_drop_write_file(filp); 659 return ret; 660 } 661 default: 662 return -ENOTTY; 663 } 664 } 665 666 #ifdef CONFIG_COMPAT 667 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 668 { 669 switch (cmd) { 670 case F2FS_IOC32_GETFLAGS: 671 cmd = F2FS_IOC_GETFLAGS; 672 break; 673 case F2FS_IOC32_SETFLAGS: 674 cmd = F2FS_IOC_SETFLAGS; 675 break; 676 default: 677 return -ENOIOCTLCMD; 678 } 679 return f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 680 } 681 #endif 682 683 const struct file_operations f2fs_file_operations = { 684 .llseek = generic_file_llseek, 685 .read = do_sync_read, 686 .write = do_sync_write, 687 .aio_read = generic_file_aio_read, 688 .aio_write = generic_file_aio_write, 689 .open = generic_file_open, 690 .mmap = f2fs_file_mmap, 691 .fsync = f2fs_sync_file, 692 .fallocate = f2fs_fallocate, 693 .unlocked_ioctl = f2fs_ioctl, 694 #ifdef CONFIG_COMPAT 695 .compat_ioctl = f2fs_compat_ioctl, 696 #endif 697 .splice_read = generic_file_splice_read, 698 .splice_write = generic_file_splice_write, 699 }; 700