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 struct dnode_of_data dn; 37 int err; 38 39 f2fs_balance_fs(sbi); 40 41 sb_start_pagefault(inode->i_sb); 42 43 /* block allocation */ 44 f2fs_lock_op(sbi); 45 set_new_dnode(&dn, inode, NULL, NULL, 0); 46 err = f2fs_reserve_block(&dn, page->index); 47 f2fs_unlock_op(sbi); 48 if (err) 49 goto out; 50 51 file_update_time(vma->vm_file); 52 lock_page(page); 53 if (unlikely(page->mapping != inode->i_mapping || 54 page_offset(page) > i_size_read(inode) || 55 !PageUptodate(page))) { 56 unlock_page(page); 57 err = -EFAULT; 58 goto out; 59 } 60 61 /* 62 * check to see if the page is mapped already (no holes) 63 */ 64 if (PageMappedToDisk(page)) 65 goto mapped; 66 67 /* page is wholly or partially inside EOF */ 68 if (((page->index + 1) << PAGE_CACHE_SHIFT) > i_size_read(inode)) { 69 unsigned offset; 70 offset = i_size_read(inode) & ~PAGE_CACHE_MASK; 71 zero_user_segment(page, offset, PAGE_CACHE_SIZE); 72 } 73 set_page_dirty(page); 74 SetPageUptodate(page); 75 76 trace_f2fs_vm_page_mkwrite(page, DATA); 77 mapped: 78 /* fill the page */ 79 wait_on_page_writeback(page); 80 out: 81 sb_end_pagefault(inode->i_sb); 82 return block_page_mkwrite_return(err); 83 } 84 85 static const struct vm_operations_struct f2fs_file_vm_ops = { 86 .fault = filemap_fault, 87 .page_mkwrite = f2fs_vm_page_mkwrite, 88 .remap_pages = generic_file_remap_pages, 89 }; 90 91 static int get_parent_ino(struct inode *inode, nid_t *pino) 92 { 93 struct dentry *dentry; 94 95 inode = igrab(inode); 96 dentry = d_find_any_alias(inode); 97 iput(inode); 98 if (!dentry) 99 return 0; 100 101 if (update_dent_inode(inode, &dentry->d_name)) { 102 dput(dentry); 103 return 0; 104 } 105 106 *pino = parent_ino(dentry); 107 dput(dentry); 108 return 1; 109 } 110 111 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) 112 { 113 struct inode *inode = file->f_mapping->host; 114 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 115 int ret = 0; 116 bool need_cp = false; 117 struct writeback_control wbc = { 118 .sync_mode = WB_SYNC_ALL, 119 .nr_to_write = LONG_MAX, 120 .for_reclaim = 0, 121 }; 122 123 if (unlikely(f2fs_readonly(inode->i_sb))) 124 return 0; 125 126 trace_f2fs_sync_file_enter(inode); 127 ret = filemap_write_and_wait_range(inode->i_mapping, start, end); 128 if (ret) { 129 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret); 130 return ret; 131 } 132 133 /* guarantee free sections for fsync */ 134 f2fs_balance_fs(sbi); 135 136 mutex_lock(&inode->i_mutex); 137 138 /* 139 * Both of fdatasync() and fsync() are able to be recovered from 140 * sudden-power-off. 141 */ 142 if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1) 143 need_cp = true; 144 else if (file_wrong_pino(inode)) 145 need_cp = true; 146 else if (!space_for_roll_forward(sbi)) 147 need_cp = true; 148 else if (!is_checkpointed_node(sbi, F2FS_I(inode)->i_pino)) 149 need_cp = true; 150 else if (F2FS_I(inode)->xattr_ver == cur_cp_version(F2FS_CKPT(sbi))) 151 need_cp = true; 152 153 if (need_cp) { 154 nid_t pino; 155 156 F2FS_I(inode)->xattr_ver = 0; 157 158 /* all the dirty node pages should be flushed for POR */ 159 ret = f2fs_sync_fs(inode->i_sb, 1); 160 if (file_wrong_pino(inode) && inode->i_nlink == 1 && 161 get_parent_ino(inode, &pino)) { 162 F2FS_I(inode)->i_pino = pino; 163 file_got_pino(inode); 164 mark_inode_dirty_sync(inode); 165 ret = f2fs_write_inode(inode, NULL); 166 if (ret) 167 goto out; 168 } 169 } else { 170 /* if there is no written node page, write its inode page */ 171 while (!sync_node_pages(sbi, inode->i_ino, &wbc)) { 172 mark_inode_dirty_sync(inode); 173 ret = f2fs_write_inode(inode, NULL); 174 if (ret) 175 goto out; 176 } 177 ret = wait_on_node_pages_writeback(sbi, inode->i_ino); 178 if (ret) 179 goto out; 180 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); 181 } 182 out: 183 mutex_unlock(&inode->i_mutex); 184 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret); 185 return ret; 186 } 187 188 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma) 189 { 190 file_accessed(file); 191 vma->vm_ops = &f2fs_file_vm_ops; 192 return 0; 193 } 194 195 int truncate_data_blocks_range(struct dnode_of_data *dn, int count) 196 { 197 int nr_free = 0, ofs = dn->ofs_in_node; 198 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb); 199 struct f2fs_node *raw_node; 200 __le32 *addr; 201 202 raw_node = F2FS_NODE(dn->node_page); 203 addr = blkaddr_in_node(raw_node) + ofs; 204 205 for ( ; count > 0; count--, addr++, dn->ofs_in_node++) { 206 block_t blkaddr = le32_to_cpu(*addr); 207 if (blkaddr == NULL_ADDR) 208 continue; 209 210 update_extent_cache(NULL_ADDR, dn); 211 invalidate_blocks(sbi, blkaddr); 212 nr_free++; 213 } 214 if (nr_free) { 215 dec_valid_block_count(sbi, dn->inode, nr_free); 216 set_page_dirty(dn->node_page); 217 sync_inode_page(dn); 218 } 219 dn->ofs_in_node = ofs; 220 221 trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid, 222 dn->ofs_in_node, nr_free); 223 return nr_free; 224 } 225 226 void truncate_data_blocks(struct dnode_of_data *dn) 227 { 228 truncate_data_blocks_range(dn, ADDRS_PER_BLOCK); 229 } 230 231 static void truncate_partial_data_page(struct inode *inode, u64 from) 232 { 233 unsigned offset = from & (PAGE_CACHE_SIZE - 1); 234 struct page *page; 235 236 if (!offset) 237 return; 238 239 page = find_data_page(inode, from >> PAGE_CACHE_SHIFT, false); 240 if (IS_ERR(page)) 241 return; 242 243 lock_page(page); 244 if (unlikely(page->mapping != inode->i_mapping)) { 245 f2fs_put_page(page, 1); 246 return; 247 } 248 wait_on_page_writeback(page); 249 zero_user(page, offset, PAGE_CACHE_SIZE - offset); 250 set_page_dirty(page); 251 f2fs_put_page(page, 1); 252 } 253 254 static int truncate_blocks(struct inode *inode, u64 from) 255 { 256 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 257 unsigned int blocksize = inode->i_sb->s_blocksize; 258 struct dnode_of_data dn; 259 pgoff_t free_from; 260 int count = 0; 261 int err; 262 263 trace_f2fs_truncate_blocks_enter(inode, from); 264 265 free_from = (pgoff_t) 266 ((from + blocksize - 1) >> (sbi->log_blocksize)); 267 268 f2fs_lock_op(sbi); 269 set_new_dnode(&dn, inode, NULL, NULL, 0); 270 err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE); 271 if (err) { 272 if (err == -ENOENT) 273 goto free_next; 274 f2fs_unlock_op(sbi); 275 trace_f2fs_truncate_blocks_exit(inode, err); 276 return err; 277 } 278 279 if (IS_INODE(dn.node_page)) 280 count = ADDRS_PER_INODE(F2FS_I(inode)); 281 else 282 count = ADDRS_PER_BLOCK; 283 284 count -= dn.ofs_in_node; 285 f2fs_bug_on(count < 0); 286 287 if (dn.ofs_in_node || IS_INODE(dn.node_page)) { 288 truncate_data_blocks_range(&dn, count); 289 free_from += count; 290 } 291 292 f2fs_put_dnode(&dn); 293 free_next: 294 err = truncate_inode_blocks(inode, free_from); 295 f2fs_unlock_op(sbi); 296 297 /* lastly zero out the first data page */ 298 truncate_partial_data_page(inode, from); 299 300 trace_f2fs_truncate_blocks_exit(inode, err); 301 return err; 302 } 303 304 void f2fs_truncate(struct inode *inode) 305 { 306 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 307 S_ISLNK(inode->i_mode))) 308 return; 309 310 trace_f2fs_truncate(inode); 311 312 if (!truncate_blocks(inode, i_size_read(inode))) { 313 inode->i_mtime = inode->i_ctime = CURRENT_TIME; 314 mark_inode_dirty(inode); 315 } 316 } 317 318 int f2fs_getattr(struct vfsmount *mnt, 319 struct dentry *dentry, struct kstat *stat) 320 { 321 struct inode *inode = dentry->d_inode; 322 generic_fillattr(inode, stat); 323 stat->blocks <<= 3; 324 return 0; 325 } 326 327 #ifdef CONFIG_F2FS_FS_POSIX_ACL 328 static void __setattr_copy(struct inode *inode, const struct iattr *attr) 329 { 330 struct f2fs_inode_info *fi = F2FS_I(inode); 331 unsigned int ia_valid = attr->ia_valid; 332 333 if (ia_valid & ATTR_UID) 334 inode->i_uid = attr->ia_uid; 335 if (ia_valid & ATTR_GID) 336 inode->i_gid = attr->ia_gid; 337 if (ia_valid & ATTR_ATIME) 338 inode->i_atime = timespec_trunc(attr->ia_atime, 339 inode->i_sb->s_time_gran); 340 if (ia_valid & ATTR_MTIME) 341 inode->i_mtime = timespec_trunc(attr->ia_mtime, 342 inode->i_sb->s_time_gran); 343 if (ia_valid & ATTR_CTIME) 344 inode->i_ctime = timespec_trunc(attr->ia_ctime, 345 inode->i_sb->s_time_gran); 346 if (ia_valid & ATTR_MODE) { 347 umode_t mode = attr->ia_mode; 348 349 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) 350 mode &= ~S_ISGID; 351 set_acl_inode(fi, mode); 352 } 353 } 354 #else 355 #define __setattr_copy setattr_copy 356 #endif 357 358 int f2fs_setattr(struct dentry *dentry, struct iattr *attr) 359 { 360 struct inode *inode = dentry->d_inode; 361 struct f2fs_inode_info *fi = F2FS_I(inode); 362 int err; 363 364 err = inode_change_ok(inode, attr); 365 if (err) 366 return err; 367 368 if ((attr->ia_valid & ATTR_SIZE) && 369 attr->ia_size != i_size_read(inode)) { 370 truncate_setsize(inode, attr->ia_size); 371 f2fs_truncate(inode); 372 f2fs_balance_fs(F2FS_SB(inode->i_sb)); 373 } 374 375 __setattr_copy(inode, attr); 376 377 if (attr->ia_valid & ATTR_MODE) { 378 err = f2fs_acl_chmod(inode); 379 if (err || is_inode_flag_set(fi, FI_ACL_MODE)) { 380 inode->i_mode = fi->i_acl_mode; 381 clear_inode_flag(fi, FI_ACL_MODE); 382 } 383 } 384 385 mark_inode_dirty(inode); 386 return err; 387 } 388 389 const struct inode_operations f2fs_file_inode_operations = { 390 .getattr = f2fs_getattr, 391 .setattr = f2fs_setattr, 392 .get_acl = f2fs_get_acl, 393 #ifdef CONFIG_F2FS_FS_XATTR 394 .setxattr = generic_setxattr, 395 .getxattr = generic_getxattr, 396 .listxattr = f2fs_listxattr, 397 .removexattr = generic_removexattr, 398 #endif 399 }; 400 401 static void fill_zero(struct inode *inode, pgoff_t index, 402 loff_t start, loff_t len) 403 { 404 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 405 struct page *page; 406 407 if (!len) 408 return; 409 410 f2fs_balance_fs(sbi); 411 412 f2fs_lock_op(sbi); 413 page = get_new_data_page(inode, NULL, index, false); 414 f2fs_unlock_op(sbi); 415 416 if (!IS_ERR(page)) { 417 wait_on_page_writeback(page); 418 zero_user(page, start, len); 419 set_page_dirty(page); 420 f2fs_put_page(page, 1); 421 } 422 } 423 424 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end) 425 { 426 pgoff_t index; 427 int err; 428 429 for (index = pg_start; index < pg_end; index++) { 430 struct dnode_of_data dn; 431 432 set_new_dnode(&dn, inode, NULL, NULL, 0); 433 err = get_dnode_of_data(&dn, index, LOOKUP_NODE); 434 if (err) { 435 if (err == -ENOENT) 436 continue; 437 return err; 438 } 439 440 if (dn.data_blkaddr != NULL_ADDR) 441 truncate_data_blocks_range(&dn, 1); 442 f2fs_put_dnode(&dn); 443 } 444 return 0; 445 } 446 447 static int punch_hole(struct inode *inode, loff_t offset, loff_t len) 448 { 449 pgoff_t pg_start, pg_end; 450 loff_t off_start, off_end; 451 int ret = 0; 452 453 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT; 454 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT; 455 456 off_start = offset & (PAGE_CACHE_SIZE - 1); 457 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1); 458 459 if (pg_start == pg_end) { 460 fill_zero(inode, pg_start, off_start, 461 off_end - off_start); 462 } else { 463 if (off_start) 464 fill_zero(inode, pg_start++, off_start, 465 PAGE_CACHE_SIZE - off_start); 466 if (off_end) 467 fill_zero(inode, pg_end, 0, off_end); 468 469 if (pg_start < pg_end) { 470 struct address_space *mapping = inode->i_mapping; 471 loff_t blk_start, blk_end; 472 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 473 474 f2fs_balance_fs(sbi); 475 476 blk_start = pg_start << PAGE_CACHE_SHIFT; 477 blk_end = pg_end << PAGE_CACHE_SHIFT; 478 truncate_inode_pages_range(mapping, blk_start, 479 blk_end - 1); 480 481 f2fs_lock_op(sbi); 482 ret = truncate_hole(inode, pg_start, pg_end); 483 f2fs_unlock_op(sbi); 484 } 485 } 486 487 return ret; 488 } 489 490 static int expand_inode_data(struct inode *inode, loff_t offset, 491 loff_t len, int mode) 492 { 493 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb); 494 pgoff_t index, pg_start, pg_end; 495 loff_t new_size = i_size_read(inode); 496 loff_t off_start, off_end; 497 int ret = 0; 498 499 ret = inode_newsize_ok(inode, (len + offset)); 500 if (ret) 501 return ret; 502 503 pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT; 504 pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT; 505 506 off_start = offset & (PAGE_CACHE_SIZE - 1); 507 off_end = (offset + len) & (PAGE_CACHE_SIZE - 1); 508 509 for (index = pg_start; index <= pg_end; index++) { 510 struct dnode_of_data dn; 511 512 f2fs_lock_op(sbi); 513 set_new_dnode(&dn, inode, NULL, NULL, 0); 514 ret = f2fs_reserve_block(&dn, index); 515 f2fs_unlock_op(sbi); 516 if (ret) 517 break; 518 519 if (pg_start == pg_end) 520 new_size = offset + len; 521 else if (index == pg_start && off_start) 522 new_size = (index + 1) << PAGE_CACHE_SHIFT; 523 else if (index == pg_end) 524 new_size = (index << PAGE_CACHE_SHIFT) + off_end; 525 else 526 new_size += PAGE_CACHE_SIZE; 527 } 528 529 if (!(mode & FALLOC_FL_KEEP_SIZE) && 530 i_size_read(inode) < new_size) { 531 i_size_write(inode, new_size); 532 mark_inode_dirty(inode); 533 } 534 535 return ret; 536 } 537 538 static long f2fs_fallocate(struct file *file, int mode, 539 loff_t offset, loff_t len) 540 { 541 struct inode *inode = file_inode(file); 542 long ret; 543 544 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 545 return -EOPNOTSUPP; 546 547 if (mode & FALLOC_FL_PUNCH_HOLE) 548 ret = punch_hole(inode, offset, len); 549 else 550 ret = expand_inode_data(inode, offset, len, mode); 551 552 if (!ret) { 553 inode->i_mtime = inode->i_ctime = CURRENT_TIME; 554 mark_inode_dirty(inode); 555 } 556 trace_f2fs_fallocate(inode, mode, offset, len, ret); 557 return ret; 558 } 559 560 #define F2FS_REG_FLMASK (~(FS_DIRSYNC_FL | FS_TOPDIR_FL)) 561 #define F2FS_OTHER_FLMASK (FS_NODUMP_FL | FS_NOATIME_FL) 562 563 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags) 564 { 565 if (S_ISDIR(mode)) 566 return flags; 567 else if (S_ISREG(mode)) 568 return flags & F2FS_REG_FLMASK; 569 else 570 return flags & F2FS_OTHER_FLMASK; 571 } 572 573 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 574 { 575 struct inode *inode = file_inode(filp); 576 struct f2fs_inode_info *fi = F2FS_I(inode); 577 unsigned int flags; 578 int ret; 579 580 switch (cmd) { 581 case F2FS_IOC_GETFLAGS: 582 flags = fi->i_flags & FS_FL_USER_VISIBLE; 583 return put_user(flags, (int __user *) arg); 584 case F2FS_IOC_SETFLAGS: 585 { 586 unsigned int oldflags; 587 588 ret = mnt_want_write_file(filp); 589 if (ret) 590 return ret; 591 592 if (!inode_owner_or_capable(inode)) { 593 ret = -EACCES; 594 goto out; 595 } 596 597 if (get_user(flags, (int __user *) arg)) { 598 ret = -EFAULT; 599 goto out; 600 } 601 602 flags = f2fs_mask_flags(inode->i_mode, flags); 603 604 mutex_lock(&inode->i_mutex); 605 606 oldflags = fi->i_flags; 607 608 if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 609 if (!capable(CAP_LINUX_IMMUTABLE)) { 610 mutex_unlock(&inode->i_mutex); 611 ret = -EPERM; 612 goto out; 613 } 614 } 615 616 flags = flags & FS_FL_USER_MODIFIABLE; 617 flags |= oldflags & ~FS_FL_USER_MODIFIABLE; 618 fi->i_flags = flags; 619 mutex_unlock(&inode->i_mutex); 620 621 f2fs_set_inode_flags(inode); 622 inode->i_ctime = CURRENT_TIME; 623 mark_inode_dirty(inode); 624 out: 625 mnt_drop_write_file(filp); 626 return ret; 627 } 628 default: 629 return -ENOTTY; 630 } 631 } 632 633 #ifdef CONFIG_COMPAT 634 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 635 { 636 switch (cmd) { 637 case F2FS_IOC32_GETFLAGS: 638 cmd = F2FS_IOC_GETFLAGS; 639 break; 640 case F2FS_IOC32_SETFLAGS: 641 cmd = F2FS_IOC_SETFLAGS; 642 break; 643 default: 644 return -ENOIOCTLCMD; 645 } 646 return f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 647 } 648 #endif 649 650 const struct file_operations f2fs_file_operations = { 651 .llseek = generic_file_llseek, 652 .read = do_sync_read, 653 .write = do_sync_write, 654 .aio_read = generic_file_aio_read, 655 .aio_write = generic_file_aio_write, 656 .open = generic_file_open, 657 .mmap = f2fs_file_mmap, 658 .fsync = f2fs_sync_file, 659 .fallocate = f2fs_fallocate, 660 .unlocked_ioctl = f2fs_ioctl, 661 #ifdef CONFIG_COMPAT 662 .compat_ioctl = f2fs_compat_ioctl, 663 #endif 664 .splice_read = generic_file_splice_read, 665 .splice_write = generic_file_splice_write, 666 }; 667