1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/f2fs/file.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8 #include <linux/blk-crypto.h> 9 #include <linux/fs.h> 10 #include <linux/f2fs_fs.h> 11 #include <linux/stat.h> 12 #include <linux/writeback.h> 13 #include <linux/blkdev.h> 14 #include <linux/falloc.h> 15 #include <linux/filelock.h> 16 #include <linux/types.h> 17 #include <linux/compat.h> 18 #include <linux/uaccess.h> 19 #include <linux/mount.h> 20 #include <linux/uio.h> 21 #include <linux/uuid.h> 22 #include <linux/file.h> 23 #include <linux/nls.h> 24 #include <linux/sched/signal.h> 25 #include <linux/fileattr.h> 26 #include <linux/fadvise.h> 27 #include <linux/iomap.h> 28 29 #include "f2fs.h" 30 #include "node.h" 31 #include "segment.h" 32 #include "xattr.h" 33 #include "acl.h" 34 #include "gc.h" 35 #include "iostat.h" 36 #include <trace/events/f2fs.h> 37 #include <uapi/linux/f2fs.h> 38 39 static void f2fs_zero_post_eof_page(struct inode *inode, 40 loff_t new_size, bool lock) 41 { 42 loff_t old_size = i_size_read(inode); 43 44 if (old_size >= new_size) 45 return; 46 47 if (mapping_empty(inode->i_mapping)) 48 return; 49 50 if (lock) 51 filemap_invalidate_lock(inode->i_mapping); 52 /* zero or drop pages only in range of [old_size, new_size] */ 53 truncate_inode_pages_range(inode->i_mapping, old_size, new_size); 54 if (lock) 55 filemap_invalidate_unlock(inode->i_mapping); 56 } 57 58 static vm_fault_t f2fs_filemap_fault(struct vm_fault *vmf) 59 { 60 struct inode *inode = file_inode(vmf->vma->vm_file); 61 vm_flags_t flags = vmf->vma->vm_flags; 62 vm_fault_t ret; 63 64 ret = filemap_fault(vmf); 65 if (ret & VM_FAULT_LOCKED) 66 f2fs_update_iostat(F2FS_I_SB(inode), inode, 67 APP_MAPPED_READ_IO, F2FS_BLKSIZE); 68 69 trace_f2fs_filemap_fault(inode, vmf->pgoff, flags, ret); 70 71 return ret; 72 } 73 74 static vm_fault_t f2fs_vm_page_mkwrite(struct vm_fault *vmf) 75 { 76 struct folio *folio = page_folio(vmf->page); 77 struct inode *inode = file_inode(vmf->vma->vm_file); 78 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 79 struct dnode_of_data dn; 80 bool need_alloc = !f2fs_is_pinned_file(inode); 81 int err = 0; 82 vm_fault_t ret; 83 84 /* 85 * We only support large folio on the read case. 86 * Don't make any dirty pages. 87 */ 88 if (unlikely(IS_IMMUTABLE(inode)) || 89 mapping_large_folio_support(inode->i_mapping)) { 90 f2fs_err(sbi, "Not expected: immutable: %d large_folio: %d", 91 IS_IMMUTABLE(inode), 92 mapping_large_folio_support(inode->i_mapping)); 93 return VM_FAULT_SIGBUS; 94 } 95 96 if (is_inode_flag_set(inode, FI_COMPRESS_RELEASED)) { 97 err = -EIO; 98 goto out; 99 } 100 101 if (unlikely(f2fs_cp_error(sbi))) { 102 err = -EIO; 103 goto out; 104 } 105 106 if (!f2fs_is_checkpoint_ready(sbi)) { 107 err = -ENOSPC; 108 goto out; 109 } 110 111 err = f2fs_convert_inline_inode(inode); 112 if (err) 113 goto out; 114 115 #ifdef CONFIG_F2FS_FS_COMPRESSION 116 if (f2fs_compressed_file(inode)) { 117 int ret = f2fs_is_compressed_cluster(inode, folio->index); 118 119 if (ret < 0) { 120 err = ret; 121 goto out; 122 } else if (ret) { 123 need_alloc = false; 124 } 125 } 126 #endif 127 /* should do out of any locked page */ 128 if (need_alloc) 129 f2fs_balance_fs(sbi, true); 130 131 sb_start_pagefault(inode->i_sb); 132 133 f2fs_bug_on(sbi, f2fs_has_inline_data(inode)); 134 135 f2fs_zero_post_eof_page(inode, (folio->index + 1) << PAGE_SHIFT, true); 136 137 file_update_time(vmf->vma->vm_file); 138 filemap_invalidate_lock_shared(inode->i_mapping); 139 140 folio_lock(folio); 141 if (unlikely(folio->mapping != inode->i_mapping || 142 folio_pos(folio) > i_size_read(inode) || 143 !folio_test_uptodate(folio))) { 144 folio_unlock(folio); 145 err = -EFAULT; 146 goto out_sem; 147 } 148 149 set_new_dnode(&dn, inode, NULL, NULL, 0); 150 if (need_alloc) { 151 /* block allocation */ 152 err = f2fs_get_block_locked(&dn, folio->index); 153 } else { 154 err = f2fs_get_dnode_of_data(&dn, folio->index, LOOKUP_NODE); 155 f2fs_put_dnode(&dn); 156 if (f2fs_is_pinned_file(inode) && 157 !__is_valid_data_blkaddr(dn.data_blkaddr)) 158 err = -EIO; 159 } 160 161 if (err) { 162 folio_unlock(folio); 163 goto out_sem; 164 } 165 166 f2fs_folio_wait_writeback(folio, DATA, false, true); 167 168 /* wait for GCed page writeback via META_MAPPING */ 169 f2fs_wait_on_block_writeback(inode, dn.data_blkaddr); 170 171 /* 172 * check to see if the page is mapped already (no holes) 173 */ 174 if (folio_test_mappedtodisk(folio)) 175 goto out_sem; 176 177 /* page is wholly or partially inside EOF */ 178 if (((loff_t)(folio->index + 1) << PAGE_SHIFT) > 179 i_size_read(inode)) { 180 loff_t offset; 181 182 offset = i_size_read(inode) & ~PAGE_MASK; 183 folio_zero_segment(folio, offset, folio_size(folio)); 184 } 185 folio_mark_dirty(folio); 186 187 f2fs_update_iostat(sbi, inode, APP_MAPPED_IO, F2FS_BLKSIZE); 188 f2fs_update_time(sbi, REQ_TIME); 189 190 out_sem: 191 filemap_invalidate_unlock_shared(inode->i_mapping); 192 193 sb_end_pagefault(inode->i_sb); 194 out: 195 ret = vmf_fs_error(err); 196 197 trace_f2fs_vm_page_mkwrite(inode, folio->index, vmf->vma->vm_flags, ret); 198 return ret; 199 } 200 201 static const struct vm_operations_struct f2fs_file_vm_ops = { 202 .fault = f2fs_filemap_fault, 203 .map_pages = filemap_map_pages, 204 .page_mkwrite = f2fs_vm_page_mkwrite, 205 }; 206 207 static int get_parent_ino(struct inode *inode, nid_t *pino) 208 { 209 struct dentry *dentry; 210 211 /* 212 * Make sure to get the non-deleted alias. The alias associated with 213 * the open file descriptor being fsync()'ed may be deleted already. 214 */ 215 dentry = d_find_alias(inode); 216 if (!dentry) 217 return 0; 218 219 *pino = d_parent_ino(dentry); 220 dput(dentry); 221 return 1; 222 } 223 224 static inline enum cp_reason_type need_do_checkpoint(struct inode *inode) 225 { 226 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 227 enum cp_reason_type cp_reason = CP_NO_NEEDED; 228 229 if (!S_ISREG(inode->i_mode)) 230 cp_reason = CP_NON_REGULAR; 231 else if (f2fs_compressed_file(inode)) 232 cp_reason = CP_COMPRESSED; 233 else if (inode->i_nlink != 1) 234 cp_reason = CP_HARDLINK; 235 else if (is_sbi_flag_set(sbi, SBI_NEED_CP)) 236 cp_reason = CP_SB_NEED_CP; 237 else if (file_wrong_pino(inode)) 238 cp_reason = CP_WRONG_PINO; 239 else if (!f2fs_space_for_roll_forward(sbi)) 240 cp_reason = CP_NO_SPC_ROLL; 241 else if (!f2fs_is_checkpointed_node(sbi, F2FS_I(inode)->i_pino)) 242 cp_reason = CP_NODE_NEED_CP; 243 else if (test_opt(sbi, FASTBOOT)) 244 cp_reason = CP_FASTBOOT_MODE; 245 else if (F2FS_OPTION(sbi).active_logs == 2) 246 cp_reason = CP_SPEC_LOG_NUM; 247 else if (F2FS_OPTION(sbi).fsync_mode == FSYNC_MODE_STRICT && 248 f2fs_need_dentry_mark(sbi, inode->i_ino) && 249 f2fs_exist_written_data(sbi, F2FS_I(inode)->i_pino, 250 TRANS_DIR_INO)) 251 cp_reason = CP_RECOVER_DIR; 252 else if (f2fs_exist_written_data(sbi, F2FS_I(inode)->i_pino, 253 XATTR_DIR_INO)) 254 cp_reason = CP_XATTR_DIR; 255 256 return cp_reason; 257 } 258 259 static bool need_inode_page_update(struct f2fs_sb_info *sbi, nid_t ino) 260 { 261 struct folio *i = filemap_get_folio(NODE_MAPPING(sbi), ino); 262 bool ret = false; 263 /* But we need to avoid that there are some inode updates */ 264 if ((!IS_ERR(i) && folio_test_dirty(i)) || 265 f2fs_need_inode_block_update(sbi, ino)) 266 ret = true; 267 f2fs_folio_put(i, false); 268 return ret; 269 } 270 271 static void try_to_fix_pino(struct inode *inode) 272 { 273 struct f2fs_inode_info *fi = F2FS_I(inode); 274 nid_t pino; 275 276 f2fs_down_write(&fi->i_sem); 277 if (file_wrong_pino(inode) && inode->i_nlink == 1 && 278 get_parent_ino(inode, &pino)) { 279 f2fs_i_pino_write(inode, pino); 280 file_got_pino(inode); 281 } 282 f2fs_up_write(&fi->i_sem); 283 } 284 285 static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end, 286 int datasync, bool atomic) 287 { 288 struct inode *inode = file->f_mapping->host; 289 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 290 nid_t ino = inode->i_ino; 291 int ret = 0; 292 enum cp_reason_type cp_reason = 0; 293 struct writeback_control wbc = { 294 .sync_mode = WB_SYNC_ALL, 295 .nr_to_write = LONG_MAX, 296 }; 297 unsigned int seq_id = 0; 298 299 if (unlikely(f2fs_readonly(inode->i_sb))) 300 return 0; 301 302 trace_f2fs_sync_file_enter(inode); 303 304 if (S_ISDIR(inode->i_mode)) 305 goto go_write; 306 307 /* if fdatasync is triggered, let's do in-place-update */ 308 if (datasync || get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks) 309 set_inode_flag(inode, FI_NEED_IPU); 310 ret = file_write_and_wait_range(file, start, end); 311 clear_inode_flag(inode, FI_NEED_IPU); 312 313 if (ret || is_sbi_flag_set(sbi, SBI_CP_DISABLED)) { 314 trace_f2fs_sync_file_exit(inode, cp_reason, datasync, ret); 315 return ret; 316 } 317 318 /* if the inode is dirty, let's recover all the time */ 319 if (!f2fs_skip_inode_update(inode, datasync)) { 320 f2fs_write_inode(inode, NULL); 321 goto go_write; 322 } 323 324 /* 325 * if there is no written data, don't waste time to write recovery info. 326 */ 327 if (!is_inode_flag_set(inode, FI_APPEND_WRITE) && 328 !f2fs_exist_written_data(sbi, ino, APPEND_INO)) { 329 330 /* it may call write_inode just prior to fsync */ 331 if (need_inode_page_update(sbi, ino)) 332 goto go_write; 333 334 if (is_inode_flag_set(inode, FI_UPDATE_WRITE) || 335 f2fs_exist_written_data(sbi, ino, UPDATE_INO)) 336 goto flush_out; 337 goto out; 338 } else { 339 /* 340 * for OPU case, during fsync(), node can be persisted before 341 * data when lower device doesn't support write barrier, result 342 * in data corruption after SPO. 343 * So for strict fsync mode, force to use atomic write semantics 344 * to keep write order in between data/node and last node to 345 * avoid potential data corruption. 346 */ 347 if (F2FS_OPTION(sbi).fsync_mode == 348 FSYNC_MODE_STRICT && !atomic) 349 atomic = true; 350 } 351 go_write: 352 /* 353 * Both of fdatasync() and fsync() are able to be recovered from 354 * sudden-power-off. 355 */ 356 f2fs_down_read(&F2FS_I(inode)->i_sem); 357 cp_reason = need_do_checkpoint(inode); 358 f2fs_up_read(&F2FS_I(inode)->i_sem); 359 360 if (cp_reason) { 361 /* all the dirty node pages should be flushed for POR */ 362 ret = f2fs_sync_fs(inode->i_sb, 1); 363 364 /* 365 * We've secured consistency through sync_fs. Following pino 366 * will be used only for fsynced inodes after checkpoint. 367 */ 368 try_to_fix_pino(inode); 369 clear_inode_flag(inode, FI_APPEND_WRITE); 370 clear_inode_flag(inode, FI_UPDATE_WRITE); 371 goto out; 372 } 373 sync_nodes: 374 atomic_inc(&sbi->wb_sync_req[NODE]); 375 ret = f2fs_fsync_node_pages(sbi, inode, &wbc, atomic, &seq_id); 376 atomic_dec(&sbi->wb_sync_req[NODE]); 377 if (ret) 378 goto out; 379 380 /* if cp_error was enabled, we should avoid infinite loop */ 381 if (unlikely(f2fs_cp_error(sbi))) { 382 ret = -EIO; 383 goto out; 384 } 385 386 if (f2fs_need_inode_block_update(sbi, ino)) { 387 f2fs_mark_inode_dirty_sync(inode, true); 388 f2fs_write_inode(inode, NULL); 389 goto sync_nodes; 390 } 391 392 /* 393 * If it's atomic_write, it's just fine to keep write ordering. So 394 * here we don't need to wait for node write completion, since we use 395 * node chain which serializes node blocks. If one of node writes are 396 * reordered, we can see simply broken chain, resulting in stopping 397 * roll-forward recovery. It means we'll recover all or none node blocks 398 * given fsync mark. 399 */ 400 if (!atomic) { 401 ret = f2fs_wait_on_node_pages_writeback(sbi, seq_id); 402 if (ret) 403 goto out; 404 } 405 406 /* once recovery info is written, don't need to tack this */ 407 f2fs_remove_ino_entry(sbi, ino, APPEND_INO); 408 clear_inode_flag(inode, FI_APPEND_WRITE); 409 flush_out: 410 if (!atomic && F2FS_OPTION(sbi).fsync_mode != FSYNC_MODE_NOBARRIER) 411 ret = f2fs_issue_flush(sbi, inode->i_ino); 412 if (!ret) { 413 f2fs_remove_ino_entry(sbi, ino, UPDATE_INO); 414 clear_inode_flag(inode, FI_UPDATE_WRITE); 415 f2fs_remove_ino_entry(sbi, ino, FLUSH_INO); 416 } 417 f2fs_update_time(sbi, REQ_TIME); 418 out: 419 trace_f2fs_sync_file_exit(inode, cp_reason, datasync, ret); 420 return ret; 421 } 422 423 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) 424 { 425 if (unlikely(f2fs_cp_error(F2FS_I_SB(file_inode(file))))) 426 return -EIO; 427 return f2fs_do_sync_file(file, start, end, datasync, false); 428 } 429 430 static bool __found_offset(struct address_space *mapping, 431 struct dnode_of_data *dn, pgoff_t index, int whence) 432 { 433 block_t blkaddr = f2fs_data_blkaddr(dn); 434 struct inode *inode = mapping->host; 435 bool compressed_cluster = false; 436 437 if (f2fs_compressed_file(inode)) { 438 block_t first_blkaddr = data_blkaddr(dn->inode, dn->node_folio, 439 ALIGN_DOWN(dn->ofs_in_node, F2FS_I(inode)->i_cluster_size)); 440 441 compressed_cluster = first_blkaddr == COMPRESS_ADDR; 442 } 443 444 switch (whence) { 445 case SEEK_DATA: 446 if (__is_valid_data_blkaddr(blkaddr)) 447 return true; 448 if (blkaddr == NEW_ADDR && 449 xa_get_mark(&mapping->i_pages, index, PAGECACHE_TAG_DIRTY)) 450 return true; 451 if (compressed_cluster) 452 return true; 453 break; 454 case SEEK_HOLE: 455 if (compressed_cluster) 456 return false; 457 if (blkaddr == NULL_ADDR) 458 return true; 459 break; 460 } 461 return false; 462 } 463 464 static loff_t f2fs_seek_block(struct file *file, loff_t offset, int whence) 465 { 466 struct inode *inode = file->f_mapping->host; 467 loff_t maxbytes = F2FS_BLK_TO_BYTES(max_file_blocks(inode)); 468 struct dnode_of_data dn; 469 pgoff_t pgofs, end_offset; 470 loff_t data_ofs = offset; 471 loff_t isize; 472 int err = 0; 473 474 inode_lock_shared(inode); 475 476 isize = i_size_read(inode); 477 if (offset >= isize) 478 goto fail; 479 480 /* handle inline data case */ 481 if (f2fs_has_inline_data(inode)) { 482 if (whence == SEEK_HOLE) { 483 data_ofs = isize; 484 goto found; 485 } else if (whence == SEEK_DATA) { 486 data_ofs = offset; 487 goto found; 488 } 489 } 490 491 pgofs = (pgoff_t)(offset >> PAGE_SHIFT); 492 493 for (; data_ofs < isize; data_ofs = (loff_t)pgofs << PAGE_SHIFT) { 494 set_new_dnode(&dn, inode, NULL, NULL, 0); 495 err = f2fs_get_dnode_of_data(&dn, pgofs, LOOKUP_NODE); 496 if (err && err != -ENOENT) { 497 goto fail; 498 } else if (err == -ENOENT) { 499 /* direct node does not exists */ 500 if (whence == SEEK_DATA) { 501 pgofs = f2fs_get_next_page_offset(&dn, pgofs); 502 continue; 503 } else { 504 goto found; 505 } 506 } 507 508 end_offset = ADDRS_PER_PAGE(dn.node_folio, inode); 509 510 /* find data/hole in dnode block */ 511 for (; dn.ofs_in_node < end_offset; 512 dn.ofs_in_node++, pgofs++, 513 data_ofs = (loff_t)pgofs << PAGE_SHIFT) { 514 block_t blkaddr; 515 516 blkaddr = f2fs_data_blkaddr(&dn); 517 518 if (__is_valid_data_blkaddr(blkaddr) && 519 !f2fs_is_valid_blkaddr(F2FS_I_SB(inode), 520 blkaddr, DATA_GENERIC_ENHANCE)) { 521 f2fs_put_dnode(&dn); 522 goto fail; 523 } 524 525 if (__found_offset(file->f_mapping, &dn, 526 pgofs, whence)) { 527 f2fs_put_dnode(&dn); 528 goto found; 529 } 530 } 531 f2fs_put_dnode(&dn); 532 } 533 534 if (whence == SEEK_DATA) 535 goto fail; 536 found: 537 if (whence == SEEK_HOLE && data_ofs > isize) 538 data_ofs = isize; 539 inode_unlock_shared(inode); 540 return vfs_setpos(file, data_ofs, maxbytes); 541 fail: 542 inode_unlock_shared(inode); 543 return -ENXIO; 544 } 545 546 static loff_t f2fs_llseek(struct file *file, loff_t offset, int whence) 547 { 548 struct inode *inode = file->f_mapping->host; 549 loff_t maxbytes = F2FS_BLK_TO_BYTES(max_file_blocks(inode)); 550 551 switch (whence) { 552 case SEEK_SET: 553 case SEEK_CUR: 554 case SEEK_END: 555 return generic_file_llseek_size(file, offset, whence, 556 maxbytes, i_size_read(inode)); 557 case SEEK_DATA: 558 case SEEK_HOLE: 559 if (offset < 0) 560 return -ENXIO; 561 return f2fs_seek_block(file, offset, whence); 562 } 563 564 return -EINVAL; 565 } 566 567 static int f2fs_file_mmap_prepare(struct vm_area_desc *desc) 568 { 569 struct file *file = desc->file; 570 struct inode *inode = file_inode(file); 571 572 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 573 return -EIO; 574 575 if (!f2fs_is_compress_backend_ready(inode)) 576 return -EOPNOTSUPP; 577 578 file_accessed(file); 579 desc->vm_ops = &f2fs_file_vm_ops; 580 581 f2fs_down_read(&F2FS_I(inode)->i_sem); 582 set_inode_flag(inode, FI_MMAP_FILE); 583 f2fs_up_read(&F2FS_I(inode)->i_sem); 584 585 return 0; 586 } 587 588 static int finish_preallocate_blocks(struct inode *inode) 589 { 590 int ret = 0; 591 bool opened; 592 593 f2fs_down_read(&F2FS_I(inode)->i_sem); 594 opened = is_inode_flag_set(inode, FI_OPENED_FILE); 595 f2fs_up_read(&F2FS_I(inode)->i_sem); 596 if (opened) 597 return 0; 598 599 inode_lock(inode); 600 if (is_inode_flag_set(inode, FI_OPENED_FILE)) 601 goto out_unlock; 602 603 if (!file_should_truncate(inode)) 604 goto out_update; 605 606 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 607 filemap_invalidate_lock(inode->i_mapping); 608 609 truncate_setsize(inode, i_size_read(inode)); 610 ret = f2fs_truncate(inode); 611 612 filemap_invalidate_unlock(inode->i_mapping); 613 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 614 if (ret) 615 goto out_unlock; 616 617 file_dont_truncate(inode); 618 out_update: 619 f2fs_down_write(&F2FS_I(inode)->i_sem); 620 set_inode_flag(inode, FI_OPENED_FILE); 621 f2fs_up_write(&F2FS_I(inode)->i_sem); 622 out_unlock: 623 inode_unlock(inode); 624 return ret; 625 } 626 627 static int f2fs_file_open(struct inode *inode, struct file *filp) 628 { 629 int err = fscrypt_file_open(inode, filp); 630 631 if (err) 632 return err; 633 634 if (!f2fs_is_compress_backend_ready(inode)) 635 return -EOPNOTSUPP; 636 637 if (mapping_large_folio_support(inode->i_mapping) && 638 filp->f_mode & FMODE_WRITE) 639 return -EOPNOTSUPP; 640 641 err = fsverity_file_open(inode, filp); 642 if (err) 643 return err; 644 645 filp->f_mode |= FMODE_NOWAIT; 646 filp->f_mode |= FMODE_CAN_ODIRECT; 647 648 err = dquot_file_open(inode, filp); 649 if (err) 650 return err; 651 652 err = finish_preallocate_blocks(inode); 653 if (!err) 654 atomic_inc(&F2FS_I(inode)->open_count); 655 return err; 656 } 657 658 void f2fs_truncate_data_blocks_range(struct dnode_of_data *dn, int count) 659 { 660 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 661 int nr_free = 0, ofs = dn->ofs_in_node, len = count; 662 __le32 *addr; 663 bool compressed_cluster = false; 664 int cluster_index = 0, valid_blocks = 0; 665 int cluster_size = F2FS_I(dn->inode)->i_cluster_size; 666 bool released = !atomic_read(&F2FS_I(dn->inode)->i_compr_blocks); 667 block_t blkstart; 668 int blklen = 0; 669 670 addr = get_dnode_addr(dn->inode, dn->node_folio) + ofs; 671 blkstart = le32_to_cpu(*addr); 672 673 /* Assumption: truncation starts with cluster */ 674 for (; count > 0; count--, addr++, dn->ofs_in_node++, cluster_index++) { 675 block_t blkaddr = le32_to_cpu(*addr); 676 677 if (f2fs_compressed_file(dn->inode) && 678 !(cluster_index & (cluster_size - 1))) { 679 if (compressed_cluster) 680 f2fs_i_compr_blocks_update(dn->inode, 681 valid_blocks, false); 682 compressed_cluster = (blkaddr == COMPRESS_ADDR); 683 valid_blocks = 0; 684 } 685 686 if (blkaddr == NULL_ADDR) 687 goto next; 688 689 f2fs_set_data_blkaddr(dn, NULL_ADDR); 690 691 if (__is_valid_data_blkaddr(blkaddr)) { 692 if (time_to_inject(sbi, FAULT_BLKADDR_CONSISTENCE)) 693 goto next; 694 if (!f2fs_is_valid_blkaddr_raw(sbi, blkaddr, 695 DATA_GENERIC_ENHANCE)) 696 goto next; 697 if (compressed_cluster) 698 valid_blocks++; 699 } 700 701 if (blkstart + blklen == blkaddr) { 702 blklen++; 703 } else { 704 f2fs_invalidate_blocks(sbi, blkstart, blklen); 705 blkstart = blkaddr; 706 blklen = 1; 707 } 708 709 if (!released || blkaddr != COMPRESS_ADDR) 710 nr_free++; 711 712 continue; 713 714 next: 715 if (blklen) 716 f2fs_invalidate_blocks(sbi, blkstart, blklen); 717 718 blkstart = le32_to_cpu(*(addr + 1)); 719 blklen = 0; 720 } 721 722 if (blklen) 723 f2fs_invalidate_blocks(sbi, blkstart, blklen); 724 725 if (compressed_cluster) 726 f2fs_i_compr_blocks_update(dn->inode, valid_blocks, false); 727 728 if (nr_free) { 729 pgoff_t fofs; 730 /* 731 * once we invalidate valid blkaddr in range [ofs, ofs + count], 732 * we will invalidate all blkaddr in the whole range. 733 */ 734 fofs = f2fs_start_bidx_of_node(ofs_of_node(dn->node_folio), 735 dn->inode) + ofs; 736 f2fs_update_read_extent_cache_range(dn, fofs, 0, len); 737 f2fs_update_age_extent_cache_range(dn, fofs, len); 738 dec_valid_block_count(sbi, dn->inode, nr_free); 739 } 740 dn->ofs_in_node = ofs; 741 742 f2fs_update_time(sbi, REQ_TIME); 743 trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid, 744 dn->ofs_in_node, nr_free); 745 } 746 747 static int truncate_partial_data_page(struct inode *inode, u64 from, 748 bool cache_only) 749 { 750 loff_t offset = from & (PAGE_SIZE - 1); 751 pgoff_t index = from >> PAGE_SHIFT; 752 struct address_space *mapping = inode->i_mapping; 753 struct folio *folio; 754 755 if (!offset && !cache_only) 756 return 0; 757 758 if (cache_only) { 759 folio = filemap_lock_folio(mapping, index); 760 if (IS_ERR(folio)) 761 return 0; 762 if (folio_test_uptodate(folio)) 763 goto truncate_out; 764 f2fs_folio_put(folio, true); 765 return 0; 766 } 767 768 folio = f2fs_get_lock_data_folio(inode, index, true); 769 if (IS_ERR(folio)) 770 return PTR_ERR(folio) == -ENOENT ? 0 : PTR_ERR(folio); 771 truncate_out: 772 f2fs_folio_wait_writeback(folio, DATA, true, true); 773 folio_zero_segment(folio, offset, folio_size(folio)); 774 775 /* An encrypted inode should have a key and truncate the last page. */ 776 f2fs_bug_on(F2FS_I_SB(inode), cache_only && IS_ENCRYPTED(inode)); 777 if (!cache_only) 778 folio_mark_dirty(folio); 779 f2fs_folio_put(folio, true); 780 return 0; 781 } 782 783 int f2fs_do_truncate_blocks(struct inode *inode, u64 from, bool lock) 784 { 785 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 786 struct dnode_of_data dn; 787 struct f2fs_lock_context lc; 788 pgoff_t free_from; 789 int count = 0, err = 0; 790 struct folio *ifolio; 791 bool truncate_page = false; 792 793 trace_f2fs_truncate_blocks_enter(inode, from); 794 795 if (IS_DEVICE_ALIASING(inode) && from) { 796 err = -EINVAL; 797 goto out_err; 798 } 799 800 free_from = (pgoff_t)F2FS_BLK_ALIGN(from); 801 802 if (free_from >= max_file_blocks(inode)) 803 goto free_partial; 804 805 if (lock) 806 f2fs_lock_op(sbi, &lc); 807 808 ifolio = f2fs_get_inode_folio(sbi, inode->i_ino); 809 if (IS_ERR(ifolio)) { 810 err = PTR_ERR(ifolio); 811 goto out; 812 } 813 814 if (IS_DEVICE_ALIASING(inode)) { 815 struct extent_tree *et = F2FS_I(inode)->extent_tree[EX_READ]; 816 struct extent_info ei = et->largest; 817 818 f2fs_invalidate_blocks(sbi, ei.blk, ei.len); 819 820 dec_valid_block_count(sbi, inode, ei.len); 821 f2fs_update_time(sbi, REQ_TIME); 822 823 f2fs_folio_put(ifolio, true); 824 goto out; 825 } 826 827 if (f2fs_has_inline_data(inode)) { 828 f2fs_truncate_inline_inode(inode, ifolio, from); 829 f2fs_folio_put(ifolio, true); 830 truncate_page = true; 831 goto out; 832 } 833 834 set_new_dnode(&dn, inode, ifolio, NULL, 0); 835 err = f2fs_get_dnode_of_data(&dn, free_from, LOOKUP_NODE_RA); 836 if (err) { 837 if (err == -ENOENT) 838 goto free_next; 839 goto out; 840 } 841 842 count = ADDRS_PER_PAGE(dn.node_folio, inode); 843 844 count -= dn.ofs_in_node; 845 f2fs_bug_on(sbi, count < 0); 846 847 if (dn.ofs_in_node || IS_INODE(dn.node_folio)) { 848 f2fs_truncate_data_blocks_range(&dn, count); 849 free_from += count; 850 } 851 852 f2fs_put_dnode(&dn); 853 free_next: 854 err = f2fs_truncate_inode_blocks(inode, free_from); 855 out: 856 if (lock) 857 f2fs_unlock_op(sbi, &lc); 858 free_partial: 859 /* lastly zero out the first data page */ 860 if (!err) 861 err = truncate_partial_data_page(inode, from, truncate_page); 862 out_err: 863 trace_f2fs_truncate_blocks_exit(inode, err); 864 return err; 865 } 866 867 int f2fs_truncate_blocks(struct inode *inode, u64 from, bool lock) 868 { 869 u64 free_from = from; 870 int err; 871 872 #ifdef CONFIG_F2FS_FS_COMPRESSION 873 /* 874 * for compressed file, only support cluster size 875 * aligned truncation. 876 */ 877 if (f2fs_compressed_file(inode)) 878 free_from = round_up(from, 879 F2FS_I(inode)->i_cluster_size << PAGE_SHIFT); 880 #endif 881 882 err = f2fs_do_truncate_blocks(inode, free_from, lock); 883 if (err) 884 return err; 885 886 #ifdef CONFIG_F2FS_FS_COMPRESSION 887 /* 888 * For compressed file, after release compress blocks, don't allow write 889 * direct, but we should allow write direct after truncate to zero. 890 */ 891 if (f2fs_compressed_file(inode) && !free_from 892 && is_inode_flag_set(inode, FI_COMPRESS_RELEASED)) 893 clear_inode_flag(inode, FI_COMPRESS_RELEASED); 894 895 if (from != free_from) { 896 err = f2fs_truncate_partial_cluster(inode, from, lock); 897 if (err) 898 return err; 899 } 900 #endif 901 902 return 0; 903 } 904 905 int f2fs_truncate(struct inode *inode) 906 { 907 int err; 908 909 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 910 return -EIO; 911 912 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 913 S_ISLNK(inode->i_mode))) 914 return 0; 915 916 trace_f2fs_truncate(inode); 917 918 if (time_to_inject(F2FS_I_SB(inode), FAULT_TRUNCATE)) 919 return -EIO; 920 921 err = f2fs_dquot_initialize(inode); 922 if (err) 923 return err; 924 925 /* we should check inline_data size */ 926 if (!f2fs_may_inline_data(inode)) { 927 err = f2fs_convert_inline_inode(inode); 928 if (err) { 929 /* 930 * Always truncate page #0 to avoid page cache 931 * leak in evict() path. 932 */ 933 truncate_inode_pages_range(inode->i_mapping, 934 F2FS_BLK_TO_BYTES(0), 935 F2FS_BLK_END_BYTES(0)); 936 return err; 937 } 938 } 939 940 err = f2fs_truncate_blocks(inode, i_size_read(inode), true); 941 if (err) 942 return err; 943 944 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 945 f2fs_mark_inode_dirty_sync(inode, false); 946 return 0; 947 } 948 949 static bool f2fs_force_buffered_io(struct inode *inode, int rw) 950 { 951 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 952 953 if (!fscrypt_dio_supported(inode)) 954 return true; 955 if (fsverity_active(inode)) 956 return true; 957 if (f2fs_compressed_file(inode)) 958 return true; 959 /* 960 * only force direct read to use buffered IO, for direct write, 961 * it expects inline data conversion before committing IO. 962 */ 963 if (f2fs_has_inline_data(inode) && rw == READ) 964 return true; 965 966 /* disallow direct IO if any of devices has unaligned blksize */ 967 if (f2fs_is_multi_device(sbi) && !sbi->aligned_blksize) 968 return true; 969 /* 970 * for blkzoned device, fallback direct IO to buffered IO, so 971 * all IOs can be serialized by log-structured write. 972 */ 973 if (f2fs_sb_has_blkzoned(sbi) && (rw == WRITE) && 974 !f2fs_is_pinned_file(inode)) 975 return true; 976 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED)) 977 return true; 978 979 return false; 980 } 981 982 int f2fs_getattr(struct mnt_idmap *idmap, const struct path *path, 983 struct kstat *stat, u32 request_mask, unsigned int query_flags) 984 { 985 struct inode *inode = d_inode(path->dentry); 986 struct f2fs_inode_info *fi = F2FS_I(inode); 987 struct f2fs_inode *ri = NULL; 988 unsigned int flags; 989 990 if (f2fs_has_extra_attr(inode) && 991 f2fs_sb_has_inode_crtime(F2FS_I_SB(inode)) && 992 F2FS_FITS_IN_INODE(ri, fi->i_extra_isize, i_crtime)) { 993 stat->result_mask |= STATX_BTIME; 994 stat->btime.tv_sec = fi->i_crtime.tv_sec; 995 stat->btime.tv_nsec = fi->i_crtime.tv_nsec; 996 } 997 998 /* 999 * Return the DIO alignment restrictions if requested. We only return 1000 * this information when requested, since on encrypted files it might 1001 * take a fair bit of work to get if the file wasn't opened recently. 1002 * 1003 * f2fs sometimes supports DIO reads but not DIO writes. STATX_DIOALIGN 1004 * cannot represent that, so in that case we report no DIO support. 1005 */ 1006 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) { 1007 unsigned int bsize = i_blocksize(inode); 1008 1009 stat->result_mask |= STATX_DIOALIGN; 1010 if (!f2fs_force_buffered_io(inode, WRITE)) { 1011 stat->dio_mem_align = bsize; 1012 stat->dio_offset_align = bsize; 1013 } 1014 } 1015 1016 flags = fi->i_flags; 1017 if (flags & F2FS_COMPR_FL) 1018 stat->attributes |= STATX_ATTR_COMPRESSED; 1019 if (flags & F2FS_APPEND_FL) 1020 stat->attributes |= STATX_ATTR_APPEND; 1021 if (IS_ENCRYPTED(inode)) 1022 stat->attributes |= STATX_ATTR_ENCRYPTED; 1023 if (flags & F2FS_IMMUTABLE_FL) 1024 stat->attributes |= STATX_ATTR_IMMUTABLE; 1025 if (flags & F2FS_NODUMP_FL) 1026 stat->attributes |= STATX_ATTR_NODUMP; 1027 if (IS_VERITY(inode)) 1028 stat->attributes |= STATX_ATTR_VERITY; 1029 1030 stat->attributes_mask |= (STATX_ATTR_COMPRESSED | 1031 STATX_ATTR_APPEND | 1032 STATX_ATTR_ENCRYPTED | 1033 STATX_ATTR_IMMUTABLE | 1034 STATX_ATTR_NODUMP | 1035 STATX_ATTR_VERITY); 1036 1037 generic_fillattr(idmap, request_mask, inode, stat); 1038 1039 /* we need to show initial sectors used for inline_data/dentries */ 1040 if ((S_ISREG(inode->i_mode) && f2fs_has_inline_data(inode)) || 1041 f2fs_has_inline_dentry(inode)) 1042 stat->blocks += (stat->size + 511) >> 9; 1043 1044 return 0; 1045 } 1046 1047 #ifdef CONFIG_F2FS_FS_POSIX_ACL 1048 static void __setattr_copy(struct mnt_idmap *idmap, 1049 struct inode *inode, const struct iattr *attr) 1050 { 1051 unsigned int ia_valid = attr->ia_valid; 1052 1053 i_uid_update(idmap, attr, inode); 1054 i_gid_update(idmap, attr, inode); 1055 if (ia_valid & ATTR_ATIME) 1056 inode_set_atime_to_ts(inode, attr->ia_atime); 1057 if (ia_valid & ATTR_MTIME) 1058 inode_set_mtime_to_ts(inode, attr->ia_mtime); 1059 if (ia_valid & ATTR_CTIME) 1060 inode_set_ctime_to_ts(inode, attr->ia_ctime); 1061 if (ia_valid & ATTR_MODE) { 1062 umode_t mode = attr->ia_mode; 1063 1064 if (!in_group_or_capable(idmap, inode, i_gid_into_vfsgid(idmap, inode))) 1065 mode &= ~S_ISGID; 1066 set_acl_inode(inode, mode); 1067 } 1068 } 1069 #else 1070 #define __setattr_copy setattr_copy 1071 #endif 1072 1073 int f2fs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 1074 struct iattr *attr) 1075 { 1076 struct inode *inode = d_inode(dentry); 1077 struct f2fs_inode_info *fi = F2FS_I(inode); 1078 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1079 int err; 1080 1081 if (unlikely(f2fs_cp_error(sbi))) 1082 return -EIO; 1083 1084 err = setattr_prepare(idmap, dentry, attr); 1085 if (err) 1086 return err; 1087 1088 err = fscrypt_prepare_setattr(dentry, attr); 1089 if (err) 1090 return err; 1091 1092 if (unlikely(IS_IMMUTABLE(inode))) 1093 return -EPERM; 1094 1095 if (unlikely(IS_APPEND(inode) && 1096 (attr->ia_valid & (ATTR_MODE | ATTR_UID | 1097 ATTR_GID | ATTR_TIMES_SET)))) 1098 return -EPERM; 1099 1100 if ((attr->ia_valid & ATTR_SIZE)) { 1101 if (!f2fs_is_compress_backend_ready(inode) || 1102 IS_DEVICE_ALIASING(inode)) 1103 return -EOPNOTSUPP; 1104 if (is_inode_flag_set(inode, FI_COMPRESS_RELEASED) && 1105 !IS_ALIGNED(attr->ia_size, 1106 F2FS_BLK_TO_BYTES(fi->i_cluster_size))) 1107 return -EINVAL; 1108 /* 1109 * To prevent scattered pin block generation, we don't allow 1110 * smaller/equal size unaligned truncation for pinned file. 1111 * We only support overwrite IO to pinned file, so don't 1112 * care about larger size truncation. 1113 */ 1114 if (f2fs_is_pinned_file(inode) && 1115 attr->ia_size <= i_size_read(inode) && 1116 !IS_ALIGNED(attr->ia_size, 1117 F2FS_BLK_TO_BYTES(CAP_BLKS_PER_SEC(sbi)))) 1118 return -EINVAL; 1119 } 1120 1121 if (is_quota_modification(idmap, inode, attr)) { 1122 err = f2fs_dquot_initialize(inode); 1123 if (err) 1124 return err; 1125 } 1126 if (i_uid_needs_update(idmap, attr, inode) || 1127 i_gid_needs_update(idmap, attr, inode)) { 1128 struct f2fs_lock_context lc; 1129 1130 f2fs_lock_op(sbi, &lc); 1131 err = dquot_transfer(idmap, inode, attr); 1132 if (err) { 1133 set_sbi_flag(sbi, SBI_QUOTA_NEED_REPAIR); 1134 f2fs_unlock_op(sbi, &lc); 1135 return err; 1136 } 1137 /* 1138 * update uid/gid under lock_op(), so that dquot and inode can 1139 * be updated atomically. 1140 */ 1141 i_uid_update(idmap, attr, inode); 1142 i_gid_update(idmap, attr, inode); 1143 f2fs_mark_inode_dirty_sync(inode, true); 1144 f2fs_unlock_op(sbi, &lc); 1145 } 1146 1147 if (attr->ia_valid & ATTR_SIZE) { 1148 loff_t old_size = i_size_read(inode); 1149 1150 if (attr->ia_size > MAX_INLINE_DATA(inode)) { 1151 /* 1152 * should convert inline inode before i_size_write to 1153 * keep smaller than inline_data size with inline flag. 1154 */ 1155 err = f2fs_convert_inline_inode(inode); 1156 if (err) 1157 return err; 1158 } 1159 1160 /* 1161 * wait for inflight dio, blocks should be removed after 1162 * IO completion. 1163 */ 1164 if (attr->ia_size < old_size) 1165 inode_dio_wait(inode); 1166 1167 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 1168 filemap_invalidate_lock(inode->i_mapping); 1169 1170 if (attr->ia_size > old_size) 1171 f2fs_zero_post_eof_page(inode, attr->ia_size, false); 1172 truncate_setsize(inode, attr->ia_size); 1173 1174 if (attr->ia_size <= old_size) 1175 err = f2fs_truncate(inode); 1176 /* 1177 * do not trim all blocks after i_size if target size is 1178 * larger than i_size. 1179 */ 1180 filemap_invalidate_unlock(inode->i_mapping); 1181 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 1182 if (err) 1183 return err; 1184 1185 spin_lock(&fi->i_size_lock); 1186 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 1187 fi->last_disk_size = i_size_read(inode); 1188 spin_unlock(&fi->i_size_lock); 1189 } 1190 1191 __setattr_copy(idmap, inode, attr); 1192 1193 if (attr->ia_valid & ATTR_MODE) { 1194 err = posix_acl_chmod(idmap, dentry, f2fs_get_inode_mode(inode)); 1195 1196 if (is_inode_flag_set(inode, FI_ACL_MODE)) { 1197 if (!err) 1198 inode->i_mode = fi->i_acl_mode; 1199 clear_inode_flag(inode, FI_ACL_MODE); 1200 } 1201 } 1202 1203 /* file size may changed here */ 1204 f2fs_mark_inode_dirty_sync(inode, true); 1205 1206 /* inode change will produce dirty node pages flushed by checkpoint */ 1207 f2fs_balance_fs(sbi, true); 1208 1209 return err; 1210 } 1211 1212 const struct inode_operations f2fs_file_inode_operations = { 1213 .getattr = f2fs_getattr, 1214 .setattr = f2fs_setattr, 1215 .get_inode_acl = f2fs_get_acl, 1216 .set_acl = f2fs_set_acl, 1217 .listxattr = f2fs_listxattr, 1218 .fiemap = f2fs_fiemap, 1219 .fileattr_get = f2fs_fileattr_get, 1220 .fileattr_set = f2fs_fileattr_set, 1221 }; 1222 1223 static int fill_zero(struct inode *inode, pgoff_t index, 1224 loff_t start, loff_t len) 1225 { 1226 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1227 struct folio *folio; 1228 struct f2fs_lock_context lc; 1229 1230 if (!len) 1231 return 0; 1232 1233 f2fs_balance_fs(sbi, true); 1234 1235 f2fs_lock_op(sbi, &lc); 1236 folio = f2fs_get_new_data_folio(inode, NULL, index, false); 1237 f2fs_unlock_op(sbi, &lc); 1238 1239 if (IS_ERR(folio)) 1240 return PTR_ERR(folio); 1241 1242 f2fs_folio_wait_writeback(folio, DATA, true, true); 1243 folio_zero_range(folio, start, len); 1244 folio_mark_dirty(folio); 1245 f2fs_folio_put(folio, true); 1246 return 0; 1247 } 1248 1249 int f2fs_truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end) 1250 { 1251 int err; 1252 1253 while (pg_start < pg_end) { 1254 struct dnode_of_data dn; 1255 pgoff_t end_offset, count; 1256 1257 set_new_dnode(&dn, inode, NULL, NULL, 0); 1258 err = f2fs_get_dnode_of_data(&dn, pg_start, LOOKUP_NODE); 1259 if (err) { 1260 if (err == -ENOENT) { 1261 pg_start = f2fs_get_next_page_offset(&dn, 1262 pg_start); 1263 continue; 1264 } 1265 return err; 1266 } 1267 1268 end_offset = ADDRS_PER_PAGE(dn.node_folio, inode); 1269 count = min(end_offset - dn.ofs_in_node, pg_end - pg_start); 1270 1271 f2fs_bug_on(F2FS_I_SB(inode), count == 0 || count > end_offset); 1272 1273 f2fs_truncate_data_blocks_range(&dn, count); 1274 f2fs_put_dnode(&dn); 1275 1276 pg_start += count; 1277 } 1278 return 0; 1279 } 1280 1281 static int f2fs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 1282 { 1283 pgoff_t pg_start, pg_end; 1284 loff_t off_start, off_end; 1285 int ret; 1286 1287 ret = f2fs_convert_inline_inode(inode); 1288 if (ret) 1289 return ret; 1290 1291 f2fs_zero_post_eof_page(inode, offset + len, true); 1292 1293 pg_start = ((unsigned long long) offset) >> PAGE_SHIFT; 1294 pg_end = ((unsigned long long) offset + len) >> PAGE_SHIFT; 1295 1296 off_start = offset & (PAGE_SIZE - 1); 1297 off_end = (offset + len) & (PAGE_SIZE - 1); 1298 1299 if (pg_start == pg_end) { 1300 ret = fill_zero(inode, pg_start, off_start, 1301 off_end - off_start); 1302 if (ret) 1303 return ret; 1304 } else { 1305 if (off_start) { 1306 ret = fill_zero(inode, pg_start++, off_start, 1307 PAGE_SIZE - off_start); 1308 if (ret) 1309 return ret; 1310 } 1311 if (off_end) { 1312 ret = fill_zero(inode, pg_end, 0, off_end); 1313 if (ret) 1314 return ret; 1315 } 1316 1317 if (pg_start < pg_end) { 1318 loff_t blk_start, blk_end; 1319 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1320 struct f2fs_lock_context lc; 1321 1322 f2fs_balance_fs(sbi, true); 1323 1324 blk_start = (loff_t)pg_start << PAGE_SHIFT; 1325 blk_end = (loff_t)pg_end << PAGE_SHIFT; 1326 1327 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1328 filemap_invalidate_lock(inode->i_mapping); 1329 1330 truncate_pagecache_range(inode, blk_start, blk_end - 1); 1331 1332 f2fs_lock_op(sbi, &lc); 1333 ret = f2fs_truncate_hole(inode, pg_start, pg_end); 1334 f2fs_unlock_op(sbi, &lc); 1335 1336 filemap_invalidate_unlock(inode->i_mapping); 1337 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1338 } 1339 } 1340 1341 return ret; 1342 } 1343 1344 static int __read_out_blkaddrs(struct inode *inode, block_t *blkaddr, 1345 int *do_replace, pgoff_t off, pgoff_t len) 1346 { 1347 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1348 struct dnode_of_data dn; 1349 int ret, done, i; 1350 1351 next_dnode: 1352 set_new_dnode(&dn, inode, NULL, NULL, 0); 1353 ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA); 1354 if (ret && ret != -ENOENT) { 1355 return ret; 1356 } else if (ret == -ENOENT) { 1357 if (dn.max_level == 0) 1358 return -ENOENT; 1359 done = min((pgoff_t)ADDRS_PER_BLOCK(inode) - 1360 dn.ofs_in_node, len); 1361 blkaddr += done; 1362 do_replace += done; 1363 goto next; 1364 } 1365 1366 done = min((pgoff_t)ADDRS_PER_PAGE(dn.node_folio, inode) - 1367 dn.ofs_in_node, len); 1368 for (i = 0; i < done; i++, blkaddr++, do_replace++, dn.ofs_in_node++) { 1369 *blkaddr = f2fs_data_blkaddr(&dn); 1370 1371 if (__is_valid_data_blkaddr(*blkaddr) && 1372 !f2fs_is_valid_blkaddr(sbi, *blkaddr, 1373 DATA_GENERIC_ENHANCE)) { 1374 f2fs_put_dnode(&dn); 1375 return -EFSCORRUPTED; 1376 } 1377 1378 if (!f2fs_is_checkpointed_data(sbi, *blkaddr)) { 1379 1380 if (f2fs_lfs_mode(sbi)) { 1381 f2fs_put_dnode(&dn); 1382 return -EOPNOTSUPP; 1383 } 1384 1385 /* do not invalidate this block address */ 1386 f2fs_update_data_blkaddr(&dn, NULL_ADDR); 1387 *do_replace = 1; 1388 } 1389 } 1390 f2fs_put_dnode(&dn); 1391 next: 1392 len -= done; 1393 off += done; 1394 if (len) 1395 goto next_dnode; 1396 return 0; 1397 } 1398 1399 static int __roll_back_blkaddrs(struct inode *inode, block_t *blkaddr, 1400 int *do_replace, pgoff_t off, int len) 1401 { 1402 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1403 struct dnode_of_data dn; 1404 int ret, i; 1405 1406 for (i = 0; i < len; i++, do_replace++, blkaddr++) { 1407 if (*do_replace == 0) 1408 continue; 1409 1410 set_new_dnode(&dn, inode, NULL, NULL, 0); 1411 ret = f2fs_get_dnode_of_data(&dn, off + i, LOOKUP_NODE_RA); 1412 if (ret) { 1413 dec_valid_block_count(sbi, inode, 1); 1414 f2fs_invalidate_blocks(sbi, *blkaddr, 1); 1415 } else { 1416 f2fs_update_data_blkaddr(&dn, *blkaddr); 1417 } 1418 f2fs_put_dnode(&dn); 1419 } 1420 return 0; 1421 } 1422 1423 static int __clone_blkaddrs(struct inode *src_inode, struct inode *dst_inode, 1424 block_t *blkaddr, int *do_replace, 1425 pgoff_t src, pgoff_t dst, pgoff_t len, bool full) 1426 { 1427 struct f2fs_sb_info *sbi = F2FS_I_SB(src_inode); 1428 pgoff_t i = 0; 1429 int ret; 1430 1431 while (i < len) { 1432 if (blkaddr[i] == NULL_ADDR && !full) { 1433 i++; 1434 continue; 1435 } 1436 1437 if (do_replace[i] || blkaddr[i] == NULL_ADDR) { 1438 struct dnode_of_data dn; 1439 struct node_info ni; 1440 size_t new_size; 1441 pgoff_t ilen; 1442 1443 set_new_dnode(&dn, dst_inode, NULL, NULL, 0); 1444 ret = f2fs_get_dnode_of_data(&dn, dst + i, ALLOC_NODE); 1445 if (ret) 1446 return ret; 1447 1448 ret = f2fs_get_node_info(sbi, dn.nid, &ni, false); 1449 if (ret) { 1450 f2fs_put_dnode(&dn); 1451 return ret; 1452 } 1453 1454 ilen = min((pgoff_t) 1455 ADDRS_PER_PAGE(dn.node_folio, dst_inode) - 1456 dn.ofs_in_node, len - i); 1457 do { 1458 dn.data_blkaddr = f2fs_data_blkaddr(&dn); 1459 f2fs_truncate_data_blocks_range(&dn, 1); 1460 1461 if (do_replace[i]) { 1462 f2fs_i_blocks_write(src_inode, 1463 1, false, false); 1464 f2fs_i_blocks_write(dst_inode, 1465 1, true, false); 1466 f2fs_replace_block(sbi, &dn, dn.data_blkaddr, 1467 blkaddr[i], ni.version, true, false); 1468 1469 do_replace[i] = 0; 1470 } 1471 dn.ofs_in_node++; 1472 i++; 1473 new_size = (loff_t)(dst + i) << PAGE_SHIFT; 1474 if (dst_inode->i_size < new_size) 1475 f2fs_i_size_write(dst_inode, new_size); 1476 } while (--ilen && (do_replace[i] || blkaddr[i] == NULL_ADDR)); 1477 1478 f2fs_put_dnode(&dn); 1479 } else { 1480 struct folio *fsrc, *fdst; 1481 1482 fsrc = f2fs_get_lock_data_folio(src_inode, 1483 src + i, true); 1484 if (IS_ERR(fsrc)) 1485 return PTR_ERR(fsrc); 1486 fdst = f2fs_get_new_data_folio(dst_inode, NULL, dst + i, 1487 true); 1488 if (IS_ERR(fdst)) { 1489 f2fs_folio_put(fsrc, true); 1490 return PTR_ERR(fdst); 1491 } 1492 1493 f2fs_folio_wait_writeback(fdst, DATA, true, true); 1494 1495 memcpy_folio(fdst, 0, fsrc, 0, PAGE_SIZE); 1496 folio_mark_dirty(fdst); 1497 folio_set_f2fs_gcing(fdst); 1498 f2fs_folio_put(fdst, true); 1499 f2fs_folio_put(fsrc, true); 1500 1501 ret = f2fs_truncate_hole(src_inode, 1502 src + i, src + i + 1); 1503 if (ret) 1504 return ret; 1505 i++; 1506 } 1507 } 1508 return 0; 1509 } 1510 1511 static int __exchange_data_block(struct inode *src_inode, 1512 struct inode *dst_inode, pgoff_t src, pgoff_t dst, 1513 pgoff_t len, bool full) 1514 { 1515 block_t *src_blkaddr; 1516 int *do_replace; 1517 pgoff_t olen; 1518 int ret; 1519 1520 while (len) { 1521 olen = min((pgoff_t)4 * ADDRS_PER_BLOCK(src_inode), len); 1522 1523 src_blkaddr = f2fs_kvzalloc(F2FS_I_SB(src_inode), 1524 array_size(olen, sizeof(block_t)), 1525 GFP_NOFS); 1526 if (!src_blkaddr) 1527 return -ENOMEM; 1528 1529 do_replace = f2fs_kvzalloc(F2FS_I_SB(src_inode), 1530 array_size(olen, sizeof(int)), 1531 GFP_NOFS); 1532 if (!do_replace) { 1533 kvfree(src_blkaddr); 1534 return -ENOMEM; 1535 } 1536 1537 ret = __read_out_blkaddrs(src_inode, src_blkaddr, 1538 do_replace, src, olen); 1539 if (ret) 1540 goto roll_back; 1541 1542 ret = __clone_blkaddrs(src_inode, dst_inode, src_blkaddr, 1543 do_replace, src, dst, olen, full); 1544 if (ret) 1545 goto roll_back; 1546 1547 src += olen; 1548 dst += olen; 1549 len -= olen; 1550 1551 kvfree(src_blkaddr); 1552 kvfree(do_replace); 1553 } 1554 return 0; 1555 1556 roll_back: 1557 __roll_back_blkaddrs(src_inode, src_blkaddr, do_replace, src, olen); 1558 kvfree(src_blkaddr); 1559 kvfree(do_replace); 1560 return ret; 1561 } 1562 1563 static int f2fs_do_collapse(struct inode *inode, loff_t offset, loff_t len) 1564 { 1565 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1566 struct f2fs_lock_context lc; 1567 pgoff_t nrpages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 1568 pgoff_t start = offset >> PAGE_SHIFT; 1569 pgoff_t end = (offset + len) >> PAGE_SHIFT; 1570 int ret; 1571 1572 f2fs_balance_fs(sbi, true); 1573 1574 /* avoid gc operation during block exchange */ 1575 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1576 filemap_invalidate_lock(inode->i_mapping); 1577 1578 f2fs_zero_post_eof_page(inode, offset + len, false); 1579 1580 f2fs_lock_op(sbi, &lc); 1581 f2fs_drop_extent_tree(inode); 1582 truncate_pagecache(inode, offset); 1583 ret = __exchange_data_block(inode, inode, end, start, nrpages - end, true); 1584 f2fs_unlock_op(sbi, &lc); 1585 1586 filemap_invalidate_unlock(inode->i_mapping); 1587 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1588 return ret; 1589 } 1590 1591 static int f2fs_collapse_range(struct inode *inode, loff_t offset, loff_t len) 1592 { 1593 loff_t new_size; 1594 int ret; 1595 1596 if (offset + len >= i_size_read(inode)) 1597 return -EINVAL; 1598 1599 /* collapse range should be aligned to block size of f2fs. */ 1600 if (offset & (F2FS_BLKSIZE - 1) || len & (F2FS_BLKSIZE - 1)) 1601 return -EINVAL; 1602 1603 ret = f2fs_convert_inline_inode(inode); 1604 if (ret) 1605 return ret; 1606 1607 /* write out all dirty pages from offset */ 1608 ret = filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX); 1609 if (ret) 1610 return ret; 1611 1612 ret = f2fs_do_collapse(inode, offset, len); 1613 if (ret) 1614 return ret; 1615 1616 /* write out all moved pages, if possible */ 1617 filemap_invalidate_lock(inode->i_mapping); 1618 filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX); 1619 truncate_pagecache(inode, offset); 1620 1621 new_size = i_size_read(inode) - len; 1622 ret = f2fs_truncate_blocks(inode, new_size, true); 1623 filemap_invalidate_unlock(inode->i_mapping); 1624 if (!ret) 1625 f2fs_i_size_write(inode, new_size); 1626 return ret; 1627 } 1628 1629 static int f2fs_do_zero_range(struct dnode_of_data *dn, pgoff_t start, 1630 pgoff_t end) 1631 { 1632 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 1633 pgoff_t index = start; 1634 unsigned int ofs_in_node = dn->ofs_in_node; 1635 blkcnt_t count = 0; 1636 int ret; 1637 1638 for (; index < end; index++, dn->ofs_in_node++) { 1639 if (f2fs_data_blkaddr(dn) == NULL_ADDR) 1640 count++; 1641 } 1642 1643 dn->ofs_in_node = ofs_in_node; 1644 ret = f2fs_reserve_new_blocks(dn, count); 1645 if (ret) 1646 return ret; 1647 1648 dn->ofs_in_node = ofs_in_node; 1649 for (index = start; index < end; index++, dn->ofs_in_node++) { 1650 dn->data_blkaddr = f2fs_data_blkaddr(dn); 1651 /* 1652 * f2fs_reserve_new_blocks will not guarantee entire block 1653 * allocation. 1654 */ 1655 if (dn->data_blkaddr == NULL_ADDR) { 1656 ret = -ENOSPC; 1657 break; 1658 } 1659 1660 if (dn->data_blkaddr == NEW_ADDR) 1661 continue; 1662 1663 if (!f2fs_is_valid_blkaddr(sbi, dn->data_blkaddr, 1664 DATA_GENERIC_ENHANCE)) { 1665 ret = -EFSCORRUPTED; 1666 break; 1667 } 1668 1669 f2fs_invalidate_blocks(sbi, dn->data_blkaddr, 1); 1670 f2fs_set_data_blkaddr(dn, NEW_ADDR); 1671 } 1672 1673 if (index > start) { 1674 f2fs_update_read_extent_cache_range(dn, start, 0, 1675 index - start); 1676 f2fs_update_age_extent_cache_range(dn, start, index - start); 1677 } 1678 1679 return ret; 1680 } 1681 1682 static int f2fs_zero_range(struct inode *inode, loff_t offset, loff_t len, 1683 int mode) 1684 { 1685 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1686 struct address_space *mapping = inode->i_mapping; 1687 pgoff_t index, pg_start, pg_end; 1688 loff_t new_size = i_size_read(inode); 1689 loff_t off_start, off_end; 1690 int ret = 0; 1691 1692 ret = inode_newsize_ok(inode, (len + offset)); 1693 if (ret) 1694 return ret; 1695 1696 ret = f2fs_convert_inline_inode(inode); 1697 if (ret) 1698 return ret; 1699 1700 ret = filemap_write_and_wait_range(mapping, offset, offset + len - 1); 1701 if (ret) 1702 return ret; 1703 1704 f2fs_zero_post_eof_page(inode, offset + len, true); 1705 1706 pg_start = ((unsigned long long) offset) >> PAGE_SHIFT; 1707 pg_end = ((unsigned long long) offset + len) >> PAGE_SHIFT; 1708 1709 off_start = offset & (PAGE_SIZE - 1); 1710 off_end = (offset + len) & (PAGE_SIZE - 1); 1711 1712 if (pg_start == pg_end) { 1713 ret = fill_zero(inode, pg_start, off_start, 1714 off_end - off_start); 1715 if (ret) 1716 return ret; 1717 1718 new_size = max_t(loff_t, new_size, offset + len); 1719 } else { 1720 if (off_start) { 1721 ret = fill_zero(inode, pg_start++, off_start, 1722 PAGE_SIZE - off_start); 1723 if (ret) 1724 return ret; 1725 1726 new_size = max_t(loff_t, new_size, 1727 (loff_t)pg_start << PAGE_SHIFT); 1728 } 1729 1730 for (index = pg_start; index < pg_end;) { 1731 struct dnode_of_data dn; 1732 struct f2fs_lock_context lc; 1733 unsigned int end_offset; 1734 pgoff_t end; 1735 1736 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1737 filemap_invalidate_lock(mapping); 1738 1739 truncate_pagecache_range(inode, 1740 (loff_t)index << PAGE_SHIFT, 1741 ((loff_t)pg_end << PAGE_SHIFT) - 1); 1742 1743 f2fs_lock_op(sbi, &lc); 1744 1745 set_new_dnode(&dn, inode, NULL, NULL, 0); 1746 ret = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE); 1747 if (ret) { 1748 f2fs_unlock_op(sbi, &lc); 1749 filemap_invalidate_unlock(mapping); 1750 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1751 goto out; 1752 } 1753 1754 end_offset = ADDRS_PER_PAGE(dn.node_folio, inode); 1755 end = min(pg_end, end_offset - dn.ofs_in_node + index); 1756 1757 ret = f2fs_do_zero_range(&dn, index, end); 1758 f2fs_put_dnode(&dn); 1759 1760 f2fs_unlock_op(sbi, &lc); 1761 filemap_invalidate_unlock(mapping); 1762 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1763 1764 f2fs_balance_fs(sbi, dn.node_changed); 1765 1766 if (ret) 1767 goto out; 1768 1769 index = end; 1770 new_size = max_t(loff_t, new_size, 1771 (loff_t)index << PAGE_SHIFT); 1772 } 1773 1774 if (off_end) { 1775 ret = fill_zero(inode, pg_end, 0, off_end); 1776 if (ret) 1777 goto out; 1778 1779 new_size = max_t(loff_t, new_size, offset + len); 1780 } 1781 } 1782 1783 out: 1784 if (new_size > i_size_read(inode)) { 1785 if (mode & FALLOC_FL_KEEP_SIZE) 1786 file_set_keep_isize(inode); 1787 else 1788 f2fs_i_size_write(inode, new_size); 1789 } 1790 return ret; 1791 } 1792 1793 static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len) 1794 { 1795 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1796 struct address_space *mapping = inode->i_mapping; 1797 pgoff_t nr, pg_start, pg_end, delta, idx; 1798 loff_t new_size; 1799 int ret = 0; 1800 1801 new_size = i_size_read(inode) + len; 1802 ret = inode_newsize_ok(inode, new_size); 1803 if (ret) 1804 return ret; 1805 1806 if (offset >= i_size_read(inode)) 1807 return -EINVAL; 1808 1809 /* insert range should be aligned to block size of f2fs. */ 1810 if (offset & (F2FS_BLKSIZE - 1) || len & (F2FS_BLKSIZE - 1)) 1811 return -EINVAL; 1812 1813 ret = f2fs_convert_inline_inode(inode); 1814 if (ret) 1815 return ret; 1816 1817 f2fs_balance_fs(sbi, true); 1818 1819 filemap_invalidate_lock(mapping); 1820 ret = f2fs_truncate_blocks(inode, i_size_read(inode), true); 1821 filemap_invalidate_unlock(mapping); 1822 if (ret) 1823 return ret; 1824 1825 /* write out all dirty pages from offset */ 1826 ret = filemap_write_and_wait_range(mapping, offset, LLONG_MAX); 1827 if (ret) 1828 return ret; 1829 1830 pg_start = offset >> PAGE_SHIFT; 1831 pg_end = (offset + len) >> PAGE_SHIFT; 1832 delta = pg_end - pg_start; 1833 idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 1834 1835 /* avoid gc operation during block exchange */ 1836 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1837 filemap_invalidate_lock(mapping); 1838 1839 f2fs_zero_post_eof_page(inode, offset + len, false); 1840 truncate_pagecache(inode, offset); 1841 1842 while (!ret && idx > pg_start) { 1843 struct f2fs_lock_context lc; 1844 1845 nr = idx - pg_start; 1846 if (nr > delta) 1847 nr = delta; 1848 idx -= nr; 1849 1850 f2fs_lock_op(sbi, &lc); 1851 f2fs_drop_extent_tree(inode); 1852 1853 ret = __exchange_data_block(inode, inode, idx, 1854 idx + delta, nr, false); 1855 f2fs_unlock_op(sbi, &lc); 1856 } 1857 filemap_invalidate_unlock(mapping); 1858 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 1859 if (ret) 1860 return ret; 1861 1862 /* write out all moved pages, if possible */ 1863 filemap_invalidate_lock(mapping); 1864 ret = filemap_write_and_wait_range(mapping, offset, LLONG_MAX); 1865 truncate_pagecache(inode, offset); 1866 filemap_invalidate_unlock(mapping); 1867 1868 if (!ret) 1869 f2fs_i_size_write(inode, new_size); 1870 return ret; 1871 } 1872 1873 static int f2fs_expand_inode_data(struct inode *inode, loff_t offset, 1874 loff_t len, int mode) 1875 { 1876 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1877 struct f2fs_map_blocks map = { .m_next_pgofs = NULL, 1878 .m_next_extent = NULL, .m_seg_type = NO_CHECK_TYPE, 1879 .m_may_create = true }; 1880 struct f2fs_gc_control gc_control = { .victim_segno = NULL_SEGNO, 1881 .init_gc_type = FG_GC, 1882 .should_migrate_blocks = false, 1883 .err_gc_skipped = true, 1884 .nr_free_secs = 0 }; 1885 pgoff_t pg_start, pg_end; 1886 loff_t new_size; 1887 loff_t off_end; 1888 block_t expanded = 0; 1889 int err; 1890 1891 err = inode_newsize_ok(inode, (len + offset)); 1892 if (err) 1893 return err; 1894 1895 err = f2fs_convert_inline_inode(inode); 1896 if (err) 1897 return err; 1898 1899 f2fs_zero_post_eof_page(inode, offset + len, true); 1900 1901 f2fs_balance_fs(sbi, true); 1902 1903 pg_start = ((unsigned long long)offset) >> PAGE_SHIFT; 1904 pg_end = ((unsigned long long)offset + len) >> PAGE_SHIFT; 1905 off_end = (offset + len) & (PAGE_SIZE - 1); 1906 1907 map.m_lblk = pg_start; 1908 map.m_len = pg_end - pg_start; 1909 if (off_end) 1910 map.m_len++; 1911 1912 if (!map.m_len) 1913 return 0; 1914 1915 if (f2fs_is_pinned_file(inode)) { 1916 block_t sec_blks = CAP_BLKS_PER_SEC(sbi); 1917 block_t sec_len = roundup(map.m_len, sec_blks); 1918 1919 map.m_len = sec_blks; 1920 next_alloc: 1921 f2fs_down_write(&sbi->pin_sem); 1922 1923 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 1924 if (has_not_enough_free_secs(sbi, 0, 0)) { 1925 f2fs_up_write(&sbi->pin_sem); 1926 err = -ENOSPC; 1927 f2fs_warn_ratelimited(sbi, 1928 "ino:%llu, start:%lu, end:%lu, need to trigger GC to " 1929 "reclaim enough free segment when checkpoint is enabled", 1930 inode->i_ino, pg_start, pg_end); 1931 goto out_err; 1932 } 1933 } 1934 1935 if (has_not_enough_free_secs(sbi, 0, 1936 sbi->reserved_pin_section)) { 1937 f2fs_down_write_trace(&sbi->gc_lock, &gc_control.lc); 1938 stat_inc_gc_call_count(sbi, FOREGROUND); 1939 err = f2fs_gc(sbi, &gc_control); 1940 if (err && err != -ENODATA) { 1941 f2fs_up_write(&sbi->pin_sem); 1942 goto out_err; 1943 } 1944 } 1945 1946 err = f2fs_allocate_pinning_section(sbi); 1947 if (err) { 1948 f2fs_up_write(&sbi->pin_sem); 1949 goto out_err; 1950 } 1951 1952 map.m_seg_type = CURSEG_COLD_DATA_PINNED; 1953 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_PRE_DIO); 1954 file_dont_truncate(inode); 1955 1956 f2fs_up_write(&sbi->pin_sem); 1957 1958 expanded += map.m_len; 1959 sec_len -= map.m_len; 1960 map.m_lblk += map.m_len; 1961 if (!err && sec_len) 1962 goto next_alloc; 1963 1964 map.m_len = expanded; 1965 } else { 1966 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_PRE_AIO); 1967 expanded = map.m_len; 1968 } 1969 out_err: 1970 if (err) { 1971 pgoff_t last_off; 1972 1973 if (!expanded) 1974 return err; 1975 1976 last_off = pg_start + expanded - 1; 1977 1978 /* update new size to the failed position */ 1979 new_size = (last_off == pg_end) ? offset + len : 1980 (loff_t)(last_off + 1) << PAGE_SHIFT; 1981 } else { 1982 new_size = ((loff_t)pg_end << PAGE_SHIFT) + off_end; 1983 } 1984 1985 if (new_size > i_size_read(inode)) { 1986 if (mode & FALLOC_FL_KEEP_SIZE) 1987 file_set_keep_isize(inode); 1988 else 1989 f2fs_i_size_write(inode, new_size); 1990 } 1991 1992 return err; 1993 } 1994 1995 static long f2fs_fallocate(struct file *file, int mode, 1996 loff_t offset, loff_t len) 1997 { 1998 struct inode *inode = file_inode(file); 1999 long ret = 0; 2000 2001 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 2002 return -EIO; 2003 if (!f2fs_is_checkpoint_ready(F2FS_I_SB(inode))) 2004 return -ENOSPC; 2005 if (!f2fs_is_compress_backend_ready(inode) || IS_DEVICE_ALIASING(inode)) 2006 return -EOPNOTSUPP; 2007 2008 /* f2fs only support ->fallocate for regular file */ 2009 if (!S_ISREG(inode->i_mode)) 2010 return -EINVAL; 2011 2012 if (IS_ENCRYPTED(inode) && 2013 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE))) 2014 return -EOPNOTSUPP; 2015 2016 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | 2017 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | 2018 FALLOC_FL_INSERT_RANGE)) 2019 return -EOPNOTSUPP; 2020 2021 inode_lock(inode); 2022 2023 /* 2024 * Pinned file should not support partial truncation since the block 2025 * can be used by applications. 2026 */ 2027 if ((f2fs_compressed_file(inode) || f2fs_is_pinned_file(inode)) && 2028 (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE | 2029 FALLOC_FL_ZERO_RANGE | FALLOC_FL_INSERT_RANGE))) { 2030 ret = -EOPNOTSUPP; 2031 goto out; 2032 } 2033 2034 ret = file_modified(file); 2035 if (ret) 2036 goto out; 2037 2038 /* 2039 * wait for inflight dio, blocks should be removed after IO 2040 * completion. 2041 */ 2042 inode_dio_wait(inode); 2043 2044 if (mode & FALLOC_FL_PUNCH_HOLE) { 2045 if (offset >= inode->i_size) 2046 goto out; 2047 2048 ret = f2fs_punch_hole(inode, offset, len); 2049 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { 2050 ret = f2fs_collapse_range(inode, offset, len); 2051 } else if (mode & FALLOC_FL_ZERO_RANGE) { 2052 ret = f2fs_zero_range(inode, offset, len, mode); 2053 } else if (mode & FALLOC_FL_INSERT_RANGE) { 2054 ret = f2fs_insert_range(inode, offset, len); 2055 } else { 2056 ret = f2fs_expand_inode_data(inode, offset, len, mode); 2057 } 2058 2059 if (!ret) { 2060 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 2061 f2fs_mark_inode_dirty_sync(inode, false); 2062 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 2063 } 2064 2065 out: 2066 inode_unlock(inode); 2067 2068 trace_f2fs_fallocate(inode, mode, offset, len, ret); 2069 return ret; 2070 } 2071 2072 static int f2fs_release_file(struct inode *inode, struct file *filp) 2073 { 2074 if (atomic_dec_and_test(&F2FS_I(inode)->open_count)) 2075 f2fs_remove_donate_inode(inode); 2076 2077 /* 2078 * f2fs_release_file is called at every close calls. So we should 2079 * not drop any inmemory pages by close called by other process. 2080 */ 2081 if (!(filp->f_mode & FMODE_WRITE) || 2082 atomic_read(&inode->i_writecount) != 1) 2083 return 0; 2084 2085 inode_lock(inode); 2086 f2fs_abort_atomic_write(inode, true); 2087 inode_unlock(inode); 2088 2089 return 0; 2090 } 2091 2092 static int f2fs_file_flush(struct file *file, fl_owner_t id) 2093 { 2094 struct inode *inode = file_inode(file); 2095 2096 /* 2097 * If the process doing a transaction is crashed, we should do 2098 * roll-back. Otherwise, other reader/write can see corrupted database 2099 * until all the writers close its file. Since this should be done 2100 * before dropping file lock, it needs to do in ->flush. 2101 */ 2102 if (F2FS_I(inode)->atomic_write_task == current && 2103 (current->flags & PF_EXITING)) { 2104 inode_lock(inode); 2105 f2fs_abort_atomic_write(inode, true); 2106 inode_unlock(inode); 2107 } 2108 2109 return 0; 2110 } 2111 2112 static int f2fs_setflags_common(struct inode *inode, u32 iflags, u32 mask) 2113 { 2114 struct f2fs_inode_info *fi = F2FS_I(inode); 2115 u32 masked_flags = fi->i_flags & mask; 2116 2117 /* mask can be shrunk by flags_valid selector */ 2118 iflags &= mask; 2119 2120 /* Is it quota file? Do not allow user to mess with it */ 2121 if (IS_NOQUOTA(inode)) 2122 return -EPERM; 2123 2124 if ((iflags ^ masked_flags) & F2FS_CASEFOLD_FL) { 2125 if (!f2fs_sb_has_casefold(F2FS_I_SB(inode))) 2126 return -EOPNOTSUPP; 2127 if (!f2fs_empty_dir(inode)) 2128 return -ENOTEMPTY; 2129 } 2130 2131 if (iflags & (F2FS_COMPR_FL | F2FS_NOCOMP_FL)) { 2132 if (!f2fs_sb_has_compression(F2FS_I_SB(inode))) 2133 return -EOPNOTSUPP; 2134 if ((iflags & F2FS_COMPR_FL) && (iflags & F2FS_NOCOMP_FL)) 2135 return -EINVAL; 2136 } 2137 2138 if ((iflags ^ masked_flags) & F2FS_COMPR_FL) { 2139 if (masked_flags & F2FS_COMPR_FL) { 2140 if (!f2fs_disable_compressed_file(inode)) 2141 return -EINVAL; 2142 } else { 2143 /* try to convert inline_data to support compression */ 2144 int err = f2fs_convert_inline_inode(inode); 2145 if (err) 2146 return err; 2147 2148 f2fs_down_write(&fi->i_sem); 2149 if (!f2fs_may_compress(inode) || 2150 atomic_read(&fi->writeback) || 2151 (S_ISREG(inode->i_mode) && 2152 F2FS_HAS_BLOCKS(inode))) { 2153 f2fs_up_write(&fi->i_sem); 2154 return -EINVAL; 2155 } 2156 err = set_compress_context(inode); 2157 f2fs_up_write(&fi->i_sem); 2158 2159 if (err) 2160 return err; 2161 } 2162 } 2163 2164 fi->i_flags = iflags | (fi->i_flags & ~mask); 2165 f2fs_bug_on(F2FS_I_SB(inode), (fi->i_flags & F2FS_COMPR_FL) && 2166 (fi->i_flags & F2FS_NOCOMP_FL)); 2167 2168 if (fi->i_flags & F2FS_PROJINHERIT_FL) 2169 set_inode_flag(inode, FI_PROJ_INHERIT); 2170 else 2171 clear_inode_flag(inode, FI_PROJ_INHERIT); 2172 2173 inode_set_ctime_current(inode); 2174 f2fs_set_inode_flags(inode); 2175 f2fs_mark_inode_dirty_sync(inode, true); 2176 return 0; 2177 } 2178 2179 /* FS_IOC_[GS]ETFLAGS and FS_IOC_FS[GS]ETXATTR support */ 2180 2181 /* 2182 * To make a new on-disk f2fs i_flag gettable via FS_IOC_GETFLAGS, add an entry 2183 * for it to f2fs_fsflags_map[], and add its FS_*_FL equivalent to 2184 * F2FS_GETTABLE_FS_FL. To also make it settable via FS_IOC_SETFLAGS, also add 2185 * its FS_*_FL equivalent to F2FS_SETTABLE_FS_FL. 2186 * 2187 * Translating flags to fsx_flags value used by FS_IOC_FSGETXATTR and 2188 * FS_IOC_FSSETXATTR is done by the VFS. 2189 */ 2190 2191 static const struct { 2192 u32 iflag; 2193 u32 fsflag; 2194 } f2fs_fsflags_map[] = { 2195 { F2FS_COMPR_FL, FS_COMPR_FL }, 2196 { F2FS_SYNC_FL, FS_SYNC_FL }, 2197 { F2FS_IMMUTABLE_FL, FS_IMMUTABLE_FL }, 2198 { F2FS_APPEND_FL, FS_APPEND_FL }, 2199 { F2FS_NODUMP_FL, FS_NODUMP_FL }, 2200 { F2FS_NOATIME_FL, FS_NOATIME_FL }, 2201 { F2FS_NOCOMP_FL, FS_NOCOMP_FL }, 2202 { F2FS_INDEX_FL, FS_INDEX_FL }, 2203 { F2FS_DIRSYNC_FL, FS_DIRSYNC_FL }, 2204 { F2FS_PROJINHERIT_FL, FS_PROJINHERIT_FL }, 2205 { F2FS_CASEFOLD_FL, FS_CASEFOLD_FL }, 2206 }; 2207 2208 #define F2FS_GETTABLE_FS_FL ( \ 2209 FS_COMPR_FL | \ 2210 FS_SYNC_FL | \ 2211 FS_IMMUTABLE_FL | \ 2212 FS_APPEND_FL | \ 2213 FS_NODUMP_FL | \ 2214 FS_NOATIME_FL | \ 2215 FS_NOCOMP_FL | \ 2216 FS_INDEX_FL | \ 2217 FS_DIRSYNC_FL | \ 2218 FS_PROJINHERIT_FL | \ 2219 FS_ENCRYPT_FL | \ 2220 FS_INLINE_DATA_FL | \ 2221 FS_NOCOW_FL | \ 2222 FS_VERITY_FL | \ 2223 FS_CASEFOLD_FL) 2224 2225 #define F2FS_SETTABLE_FS_FL ( \ 2226 FS_COMPR_FL | \ 2227 FS_SYNC_FL | \ 2228 FS_IMMUTABLE_FL | \ 2229 FS_APPEND_FL | \ 2230 FS_NODUMP_FL | \ 2231 FS_NOATIME_FL | \ 2232 FS_NOCOMP_FL | \ 2233 FS_DIRSYNC_FL | \ 2234 FS_PROJINHERIT_FL | \ 2235 FS_CASEFOLD_FL) 2236 2237 /* Convert f2fs on-disk i_flags to FS_IOC_{GET,SET}FLAGS flags */ 2238 static inline u32 f2fs_iflags_to_fsflags(u32 iflags) 2239 { 2240 u32 fsflags = 0; 2241 int i; 2242 2243 for (i = 0; i < ARRAY_SIZE(f2fs_fsflags_map); i++) 2244 if (iflags & f2fs_fsflags_map[i].iflag) 2245 fsflags |= f2fs_fsflags_map[i].fsflag; 2246 2247 return fsflags; 2248 } 2249 2250 /* Convert FS_IOC_{GET,SET}FLAGS flags to f2fs on-disk i_flags */ 2251 static inline u32 f2fs_fsflags_to_iflags(u32 fsflags) 2252 { 2253 u32 iflags = 0; 2254 int i; 2255 2256 for (i = 0; i < ARRAY_SIZE(f2fs_fsflags_map); i++) 2257 if (fsflags & f2fs_fsflags_map[i].fsflag) 2258 iflags |= f2fs_fsflags_map[i].iflag; 2259 2260 return iflags; 2261 } 2262 2263 static int f2fs_ioc_getversion(struct file *filp, unsigned long arg) 2264 { 2265 struct inode *inode = file_inode(filp); 2266 2267 return put_user(inode->i_generation, (int __user *)arg); 2268 } 2269 2270 static int f2fs_ioc_start_atomic_write(struct file *filp, bool truncate) 2271 { 2272 struct inode *inode = file_inode(filp); 2273 struct mnt_idmap *idmap = file_mnt_idmap(filp); 2274 struct f2fs_inode_info *fi = F2FS_I(inode); 2275 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2276 loff_t isize; 2277 int ret; 2278 2279 if (!(filp->f_mode & FMODE_WRITE)) 2280 return -EBADF; 2281 2282 if (!inode_owner_or_capable(idmap, inode)) 2283 return -EACCES; 2284 2285 if (!S_ISREG(inode->i_mode)) 2286 return -EINVAL; 2287 2288 if (filp->f_flags & O_DIRECT) 2289 return -EINVAL; 2290 2291 ret = mnt_want_write_file(filp); 2292 if (ret) 2293 return ret; 2294 2295 inode_lock(inode); 2296 2297 if (!f2fs_disable_compressed_file(inode) || 2298 f2fs_is_pinned_file(inode)) { 2299 ret = -EINVAL; 2300 goto out; 2301 } 2302 2303 if (f2fs_is_atomic_file(inode)) 2304 goto out; 2305 2306 ret = f2fs_convert_inline_inode(inode); 2307 if (ret) 2308 goto out; 2309 2310 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 2311 f2fs_down_write(&fi->i_gc_rwsem[READ]); 2312 2313 /* 2314 * Should wait end_io to count F2FS_WB_CP_DATA correctly by 2315 * f2fs_is_atomic_file. 2316 */ 2317 if (get_dirty_pages(inode)) 2318 f2fs_warn(sbi, "Unexpected flush for atomic writes: ino=%llu, npages=%u", 2319 inode->i_ino, get_dirty_pages(inode)); 2320 ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 2321 if (ret) 2322 goto out_unlock; 2323 2324 /* Check if the inode already has a COW inode */ 2325 if (fi->cow_inode == NULL) { 2326 /* Create a COW inode for atomic write */ 2327 struct dentry *dentry = file_dentry(filp); 2328 struct inode *dir = d_inode(dentry->d_parent); 2329 2330 ret = f2fs_get_tmpfile(idmap, dir, &fi->cow_inode); 2331 if (ret) 2332 goto out_unlock; 2333 2334 set_inode_flag(fi->cow_inode, FI_COW_FILE); 2335 clear_inode_flag(fi->cow_inode, FI_INLINE_DATA); 2336 2337 /* Set the COW inode's atomic_inode to the atomic inode */ 2338 F2FS_I(fi->cow_inode)->atomic_inode = inode; 2339 } else { 2340 /* Reuse the already created COW inode */ 2341 f2fs_bug_on(sbi, get_dirty_pages(fi->cow_inode)); 2342 2343 invalidate_mapping_pages(fi->cow_inode->i_mapping, 0, -1); 2344 2345 ret = f2fs_do_truncate_blocks(fi->cow_inode, 0, true); 2346 if (ret) 2347 goto out_unlock; 2348 } 2349 2350 f2fs_write_inode(inode, NULL); 2351 2352 stat_inc_atomic_inode(inode); 2353 2354 set_inode_flag(inode, FI_ATOMIC_FILE); 2355 2356 isize = i_size_read(inode); 2357 fi->original_i_size = isize; 2358 if (truncate) { 2359 set_inode_flag(inode, FI_ATOMIC_REPLACE); 2360 truncate_inode_pages_final(inode->i_mapping); 2361 f2fs_i_size_write(inode, 0); 2362 isize = 0; 2363 } 2364 f2fs_i_size_write(fi->cow_inode, isize); 2365 2366 out_unlock: 2367 f2fs_up_write(&fi->i_gc_rwsem[READ]); 2368 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 2369 if (ret) 2370 goto out; 2371 2372 f2fs_update_time(sbi, REQ_TIME); 2373 fi->atomic_write_task = current; 2374 stat_update_max_atomic_write(inode); 2375 fi->atomic_write_cnt = 0; 2376 out: 2377 inode_unlock(inode); 2378 mnt_drop_write_file(filp); 2379 return ret; 2380 } 2381 2382 static int f2fs_ioc_commit_atomic_write(struct file *filp) 2383 { 2384 struct inode *inode = file_inode(filp); 2385 struct mnt_idmap *idmap = file_mnt_idmap(filp); 2386 int ret; 2387 2388 if (!(filp->f_mode & FMODE_WRITE)) 2389 return -EBADF; 2390 2391 if (!inode_owner_or_capable(idmap, inode)) 2392 return -EACCES; 2393 2394 ret = mnt_want_write_file(filp); 2395 if (ret) 2396 return ret; 2397 2398 f2fs_balance_fs(F2FS_I_SB(inode), true); 2399 2400 inode_lock(inode); 2401 2402 if (f2fs_is_atomic_file(inode)) { 2403 ret = f2fs_commit_atomic_write(inode); 2404 if (!ret) 2405 ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true); 2406 2407 f2fs_abort_atomic_write(inode, ret); 2408 } else { 2409 ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 1, false); 2410 } 2411 2412 inode_unlock(inode); 2413 mnt_drop_write_file(filp); 2414 return ret; 2415 } 2416 2417 static int f2fs_ioc_abort_atomic_write(struct file *filp) 2418 { 2419 struct inode *inode = file_inode(filp); 2420 struct mnt_idmap *idmap = file_mnt_idmap(filp); 2421 int ret; 2422 2423 if (!(filp->f_mode & FMODE_WRITE)) 2424 return -EBADF; 2425 2426 if (!inode_owner_or_capable(idmap, inode)) 2427 return -EACCES; 2428 2429 ret = mnt_want_write_file(filp); 2430 if (ret) 2431 return ret; 2432 2433 inode_lock(inode); 2434 2435 f2fs_abort_atomic_write(inode, true); 2436 2437 inode_unlock(inode); 2438 2439 mnt_drop_write_file(filp); 2440 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 2441 return ret; 2442 } 2443 2444 int f2fs_do_shutdown(struct f2fs_sb_info *sbi, unsigned int flag, 2445 bool readonly, bool need_lock) 2446 { 2447 struct super_block *sb = sbi->sb; 2448 int ret = 0; 2449 2450 switch (flag) { 2451 case F2FS_GOING_DOWN_FULLSYNC: 2452 ret = bdev_freeze(sb->s_bdev); 2453 if (ret) 2454 goto out; 2455 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_SHUTDOWN); 2456 bdev_thaw(sb->s_bdev); 2457 break; 2458 case F2FS_GOING_DOWN_METASYNC: 2459 /* do checkpoint only */ 2460 ret = f2fs_sync_fs(sb, 1); 2461 if (ret) { 2462 if (ret == -EIO) 2463 ret = 0; 2464 goto out; 2465 } 2466 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_SHUTDOWN); 2467 break; 2468 case F2FS_GOING_DOWN_NOSYNC: 2469 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_SHUTDOWN); 2470 break; 2471 case F2FS_GOING_DOWN_METAFLUSH: 2472 f2fs_sync_meta_pages(sbi, LONG_MAX, FS_META_IO); 2473 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_SHUTDOWN); 2474 break; 2475 case F2FS_GOING_DOWN_NEED_FSCK: 2476 set_sbi_flag(sbi, SBI_NEED_FSCK); 2477 set_sbi_flag(sbi, SBI_CP_DISABLED_QUICK); 2478 set_sbi_flag(sbi, SBI_IS_DIRTY); 2479 /* do checkpoint only */ 2480 ret = f2fs_sync_fs(sb, 1); 2481 if (ret == -EIO) 2482 ret = 0; 2483 goto out; 2484 default: 2485 ret = -EINVAL; 2486 goto out; 2487 } 2488 2489 if (readonly) 2490 goto out; 2491 2492 /* 2493 * grab sb->s_umount to avoid racing w/ remount() and other shutdown 2494 * paths. 2495 */ 2496 if (need_lock) 2497 down_write(&sbi->sb->s_umount); 2498 2499 f2fs_stop_gc_thread(sbi); 2500 f2fs_stop_discard_thread(sbi); 2501 2502 f2fs_drop_discard_cmd(sbi); 2503 clear_opt(sbi, DISCARD); 2504 2505 if (need_lock) 2506 up_write(&sbi->sb->s_umount); 2507 2508 f2fs_update_time(sbi, REQ_TIME); 2509 out: 2510 2511 trace_f2fs_shutdown(sbi, flag, ret); 2512 2513 return ret; 2514 } 2515 2516 static int f2fs_ioc_shutdown(struct file *filp, unsigned long arg) 2517 { 2518 struct inode *inode = file_inode(filp); 2519 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2520 __u32 in; 2521 int ret; 2522 bool need_drop = false, readonly = false; 2523 2524 if (!capable(CAP_SYS_ADMIN)) 2525 return -EPERM; 2526 2527 if (get_user(in, (__u32 __user *)arg)) 2528 return -EFAULT; 2529 2530 if (in != F2FS_GOING_DOWN_FULLSYNC) { 2531 ret = mnt_want_write_file(filp); 2532 if (ret) { 2533 if (ret != -EROFS) 2534 return ret; 2535 2536 /* fallback to nosync shutdown for readonly fs */ 2537 in = F2FS_GOING_DOWN_NOSYNC; 2538 readonly = true; 2539 } else { 2540 need_drop = true; 2541 } 2542 } 2543 2544 ret = f2fs_do_shutdown(sbi, in, readonly, true); 2545 2546 if (need_drop) 2547 mnt_drop_write_file(filp); 2548 2549 return ret; 2550 } 2551 2552 static int f2fs_keep_noreuse_range(struct inode *inode, 2553 loff_t offset, loff_t len) 2554 { 2555 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2556 u64 max_bytes = F2FS_BLK_TO_BYTES(max_file_blocks(inode)); 2557 u64 start, end; 2558 int ret = 0; 2559 2560 if (!S_ISREG(inode->i_mode)) 2561 return 0; 2562 2563 if (offset >= max_bytes || len > max_bytes || 2564 (offset + len) > max_bytes) 2565 return 0; 2566 2567 start = offset >> PAGE_SHIFT; 2568 end = DIV_ROUND_UP(offset + len, PAGE_SIZE); 2569 2570 inode_lock(inode); 2571 if (f2fs_is_atomic_file(inode)) { 2572 inode_unlock(inode); 2573 return 0; 2574 } 2575 2576 spin_lock(&sbi->inode_lock[DONATE_INODE]); 2577 /* let's remove the range, if len = 0 */ 2578 if (!len) { 2579 if (!list_empty(&F2FS_I(inode)->gdonate_list)) { 2580 list_del_init(&F2FS_I(inode)->gdonate_list); 2581 sbi->donate_files--; 2582 if (is_inode_flag_set(inode, FI_DONATE_FINISHED)) 2583 ret = -EALREADY; 2584 else 2585 set_inode_flag(inode, FI_DONATE_FINISHED); 2586 } else 2587 ret = -ENOENT; 2588 } else { 2589 if (list_empty(&F2FS_I(inode)->gdonate_list)) { 2590 list_add_tail(&F2FS_I(inode)->gdonate_list, 2591 &sbi->inode_list[DONATE_INODE]); 2592 sbi->donate_files++; 2593 } else { 2594 list_move_tail(&F2FS_I(inode)->gdonate_list, 2595 &sbi->inode_list[DONATE_INODE]); 2596 } 2597 F2FS_I(inode)->donate_start = start; 2598 F2FS_I(inode)->donate_end = end - 1; 2599 clear_inode_flag(inode, FI_DONATE_FINISHED); 2600 } 2601 spin_unlock(&sbi->inode_lock[DONATE_INODE]); 2602 inode_unlock(inode); 2603 2604 return ret; 2605 } 2606 2607 static int f2fs_ioc_fitrim(struct file *filp, unsigned long arg) 2608 { 2609 struct inode *inode = file_inode(filp); 2610 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2611 struct fstrim_range range; 2612 int ret; 2613 2614 if (!capable(CAP_SYS_ADMIN)) 2615 return -EPERM; 2616 2617 if (!f2fs_hw_support_discard(sbi)) 2618 return -EOPNOTSUPP; 2619 2620 if (copy_from_user(&range, (struct fstrim_range __user *)arg, 2621 sizeof(range))) 2622 return -EFAULT; 2623 2624 ret = mnt_want_write_file(filp); 2625 if (ret) 2626 return ret; 2627 2628 range.minlen = max_t(unsigned int, range.minlen, 2629 f2fs_hw_discard_granularity(sbi)); 2630 ret = f2fs_trim_fs(sbi, &range); 2631 mnt_drop_write_file(filp); 2632 if (ret < 0) 2633 return ret; 2634 2635 if (copy_to_user((struct fstrim_range __user *)arg, &range, 2636 sizeof(range))) 2637 return -EFAULT; 2638 f2fs_update_time(sbi, REQ_TIME); 2639 return 0; 2640 } 2641 2642 static bool uuid_is_nonzero(__u8 u[16]) 2643 { 2644 int i; 2645 2646 for (i = 0; i < 16; i++) 2647 if (u[i]) 2648 return true; 2649 return false; 2650 } 2651 2652 static int f2fs_ioc_set_encryption_policy(struct file *filp, unsigned long arg) 2653 { 2654 struct inode *inode = file_inode(filp); 2655 int ret; 2656 2657 if (!f2fs_sb_has_encrypt(F2FS_I_SB(inode))) 2658 return -EOPNOTSUPP; 2659 2660 ret = fscrypt_ioctl_set_policy(filp, (const void __user *)arg); 2661 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 2662 return ret; 2663 } 2664 2665 static int f2fs_ioc_get_encryption_policy(struct file *filp, unsigned long arg) 2666 { 2667 if (!f2fs_sb_has_encrypt(F2FS_I_SB(file_inode(filp)))) 2668 return -EOPNOTSUPP; 2669 return fscrypt_ioctl_get_policy(filp, (void __user *)arg); 2670 } 2671 2672 static int f2fs_ioc_get_encryption_pwsalt(struct file *filp, unsigned long arg) 2673 { 2674 struct inode *inode = file_inode(filp); 2675 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2676 u8 encrypt_pw_salt[16]; 2677 int err; 2678 2679 if (!f2fs_sb_has_encrypt(sbi)) 2680 return -EOPNOTSUPP; 2681 2682 err = mnt_want_write_file(filp); 2683 if (err) 2684 return err; 2685 2686 f2fs_down_write(&sbi->sb_lock); 2687 2688 if (uuid_is_nonzero(sbi->raw_super->encrypt_pw_salt)) 2689 goto got_it; 2690 2691 /* update superblock with uuid */ 2692 generate_random_uuid(sbi->raw_super->encrypt_pw_salt); 2693 2694 err = f2fs_commit_super(sbi, false); 2695 if (err) { 2696 /* undo new data */ 2697 memset(sbi->raw_super->encrypt_pw_salt, 0, 16); 2698 goto out_err; 2699 } 2700 got_it: 2701 memcpy(encrypt_pw_salt, sbi->raw_super->encrypt_pw_salt, 16); 2702 out_err: 2703 f2fs_up_write(&sbi->sb_lock); 2704 mnt_drop_write_file(filp); 2705 2706 if (!err && copy_to_user((__u8 __user *)arg, encrypt_pw_salt, 16)) 2707 err = -EFAULT; 2708 2709 return err; 2710 } 2711 2712 static int f2fs_ioc_get_encryption_policy_ex(struct file *filp, 2713 unsigned long arg) 2714 { 2715 if (!f2fs_sb_has_encrypt(F2FS_I_SB(file_inode(filp)))) 2716 return -EOPNOTSUPP; 2717 2718 return fscrypt_ioctl_get_policy_ex(filp, (void __user *)arg); 2719 } 2720 2721 static int f2fs_ioc_add_encryption_key(struct file *filp, unsigned long arg) 2722 { 2723 if (!f2fs_sb_has_encrypt(F2FS_I_SB(file_inode(filp)))) 2724 return -EOPNOTSUPP; 2725 2726 return fscrypt_ioctl_add_key(filp, (void __user *)arg); 2727 } 2728 2729 static int f2fs_ioc_remove_encryption_key(struct file *filp, unsigned long arg) 2730 { 2731 if (!f2fs_sb_has_encrypt(F2FS_I_SB(file_inode(filp)))) 2732 return -EOPNOTSUPP; 2733 2734 return fscrypt_ioctl_remove_key(filp, (void __user *)arg); 2735 } 2736 2737 static int f2fs_ioc_remove_encryption_key_all_users(struct file *filp, 2738 unsigned long arg) 2739 { 2740 if (!f2fs_sb_has_encrypt(F2FS_I_SB(file_inode(filp)))) 2741 return -EOPNOTSUPP; 2742 2743 return fscrypt_ioctl_remove_key_all_users(filp, (void __user *)arg); 2744 } 2745 2746 static int f2fs_ioc_get_encryption_key_status(struct file *filp, 2747 unsigned long arg) 2748 { 2749 if (!f2fs_sb_has_encrypt(F2FS_I_SB(file_inode(filp)))) 2750 return -EOPNOTSUPP; 2751 2752 return fscrypt_ioctl_get_key_status(filp, (void __user *)arg); 2753 } 2754 2755 static int f2fs_ioc_get_encryption_nonce(struct file *filp, unsigned long arg) 2756 { 2757 if (!f2fs_sb_has_encrypt(F2FS_I_SB(file_inode(filp)))) 2758 return -EOPNOTSUPP; 2759 2760 return fscrypt_ioctl_get_nonce(filp, (void __user *)arg); 2761 } 2762 2763 static int f2fs_ioc_gc(struct file *filp, unsigned long arg) 2764 { 2765 struct inode *inode = file_inode(filp); 2766 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2767 struct f2fs_gc_control gc_control = { .victim_segno = NULL_SEGNO, 2768 .no_bg_gc = false, 2769 .should_migrate_blocks = false, 2770 .nr_free_secs = 0 }; 2771 __u32 sync; 2772 int ret; 2773 2774 if (!capable(CAP_SYS_ADMIN)) 2775 return -EPERM; 2776 2777 if (get_user(sync, (__u32 __user *)arg)) 2778 return -EFAULT; 2779 2780 if (f2fs_readonly(sbi->sb)) 2781 return -EROFS; 2782 2783 ret = mnt_want_write_file(filp); 2784 if (ret) 2785 return ret; 2786 2787 if (!sync) { 2788 if (!f2fs_down_write_trylock_trace(&sbi->gc_lock, 2789 &gc_control.lc)) { 2790 ret = -EBUSY; 2791 goto out; 2792 } 2793 } else { 2794 f2fs_down_write_trace(&sbi->gc_lock, &gc_control.lc); 2795 } 2796 2797 gc_control.init_gc_type = sync ? FG_GC : BG_GC; 2798 gc_control.err_gc_skipped = sync; 2799 stat_inc_gc_call_count(sbi, FOREGROUND); 2800 ret = f2fs_gc(sbi, &gc_control); 2801 out: 2802 mnt_drop_write_file(filp); 2803 return ret; 2804 } 2805 2806 static int __f2fs_ioc_gc_range(struct file *filp, struct f2fs_gc_range *range) 2807 { 2808 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp)); 2809 struct f2fs_gc_control gc_control = { 2810 .init_gc_type = range->sync ? FG_GC : BG_GC, 2811 .no_bg_gc = false, 2812 .should_migrate_blocks = false, 2813 .err_gc_skipped = range->sync, 2814 .nr_free_secs = 0 }; 2815 u64 end; 2816 int ret; 2817 2818 if (!capable(CAP_SYS_ADMIN)) 2819 return -EPERM; 2820 if (f2fs_readonly(sbi->sb)) 2821 return -EROFS; 2822 2823 end = range->start + range->len; 2824 if (end < range->start || range->start < MAIN_BLKADDR(sbi) || 2825 end >= MAX_BLKADDR(sbi)) 2826 return -EINVAL; 2827 2828 ret = mnt_want_write_file(filp); 2829 if (ret) 2830 return ret; 2831 2832 do_more: 2833 if (!range->sync) { 2834 if (!f2fs_down_write_trylock_trace(&sbi->gc_lock, &gc_control.lc)) { 2835 ret = -EBUSY; 2836 goto out; 2837 } 2838 } else { 2839 f2fs_down_write_trace(&sbi->gc_lock, &gc_control.lc); 2840 } 2841 2842 gc_control.victim_segno = GET_SEGNO(sbi, range->start); 2843 stat_inc_gc_call_count(sbi, FOREGROUND); 2844 ret = f2fs_gc(sbi, &gc_control); 2845 if (ret) { 2846 if (ret == -EBUSY) 2847 ret = -EAGAIN; 2848 goto out; 2849 } 2850 range->start += CAP_BLKS_PER_SEC(sbi); 2851 if (range->start <= end) 2852 goto do_more; 2853 out: 2854 mnt_drop_write_file(filp); 2855 return ret; 2856 } 2857 2858 static int f2fs_ioc_gc_range(struct file *filp, unsigned long arg) 2859 { 2860 struct f2fs_gc_range range; 2861 2862 if (copy_from_user(&range, (struct f2fs_gc_range __user *)arg, 2863 sizeof(range))) 2864 return -EFAULT; 2865 return __f2fs_ioc_gc_range(filp, &range); 2866 } 2867 2868 static int f2fs_ioc_write_checkpoint(struct file *filp) 2869 { 2870 struct inode *inode = file_inode(filp); 2871 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 2872 int ret; 2873 2874 if (!capable(CAP_SYS_ADMIN)) 2875 return -EPERM; 2876 2877 if (f2fs_readonly(sbi->sb)) 2878 return -EROFS; 2879 2880 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) { 2881 f2fs_info(sbi, "Skipping Checkpoint. Checkpoints currently disabled."); 2882 return -EINVAL; 2883 } 2884 2885 ret = mnt_want_write_file(filp); 2886 if (ret) 2887 return ret; 2888 2889 ret = f2fs_sync_fs(sbi->sb, 1); 2890 2891 mnt_drop_write_file(filp); 2892 return ret; 2893 } 2894 2895 static int f2fs_defragment_range(struct f2fs_sb_info *sbi, 2896 struct file *filp, 2897 struct f2fs_defragment *range) 2898 { 2899 struct inode *inode = file_inode(filp); 2900 struct f2fs_map_blocks map = { .m_next_extent = NULL, 2901 .m_seg_type = NO_CHECK_TYPE, 2902 .m_may_create = false }; 2903 struct extent_info ei = {}; 2904 pgoff_t pg_start, pg_end, next_pgofs; 2905 unsigned int total = 0, sec_num; 2906 block_t blk_end = 0; 2907 bool fragmented = false; 2908 int err; 2909 2910 f2fs_balance_fs(sbi, true); 2911 2912 inode_lock(inode); 2913 pg_start = range->start >> PAGE_SHIFT; 2914 pg_end = min_t(pgoff_t, 2915 (range->start + range->len) >> PAGE_SHIFT, 2916 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE)); 2917 2918 if (is_inode_flag_set(inode, FI_COMPRESS_RELEASED) || 2919 f2fs_is_atomic_file(inode)) { 2920 err = -EINVAL; 2921 goto unlock_out; 2922 } 2923 2924 /* if in-place-update policy is enabled, don't waste time here */ 2925 set_inode_flag(inode, FI_OPU_WRITE); 2926 if (f2fs_should_update_inplace(inode, NULL)) { 2927 err = -EINVAL; 2928 goto out; 2929 } 2930 2931 /* writeback all dirty pages in the range */ 2932 err = filemap_write_and_wait_range(inode->i_mapping, 2933 pg_start << PAGE_SHIFT, 2934 (pg_end << PAGE_SHIFT) - 1); 2935 if (err) 2936 goto out; 2937 2938 /* 2939 * lookup mapping info in extent cache, skip defragmenting if physical 2940 * block addresses are continuous. 2941 */ 2942 if (f2fs_lookup_read_extent_cache(inode, pg_start, &ei)) { 2943 if ((pgoff_t)ei.fofs + ei.len >= pg_end) 2944 goto out; 2945 } 2946 2947 map.m_lblk = pg_start; 2948 map.m_next_pgofs = &next_pgofs; 2949 2950 /* 2951 * lookup mapping info in dnode page cache, skip defragmenting if all 2952 * physical block addresses are continuous even if there are hole(s) 2953 * in logical blocks. 2954 */ 2955 while (map.m_lblk < pg_end) { 2956 map.m_len = pg_end - map.m_lblk; 2957 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT); 2958 if (err) 2959 goto out; 2960 2961 if (!(map.m_flags & F2FS_MAP_FLAGS)) { 2962 map.m_lblk = next_pgofs; 2963 continue; 2964 } 2965 2966 if (blk_end && blk_end != map.m_pblk) 2967 fragmented = true; 2968 2969 /* record total count of block that we're going to move */ 2970 total += map.m_len; 2971 2972 blk_end = map.m_pblk + map.m_len; 2973 2974 map.m_lblk += map.m_len; 2975 } 2976 2977 if (!fragmented) { 2978 total = 0; 2979 goto out; 2980 } 2981 2982 sec_num = DIV_ROUND_UP(total, CAP_BLKS_PER_SEC(sbi)); 2983 2984 /* 2985 * make sure there are enough free section for LFS allocation, this can 2986 * avoid defragment running in SSR mode when free section are allocated 2987 * intensively 2988 */ 2989 if (has_not_enough_free_secs(sbi, 0, sec_num)) { 2990 err = -EAGAIN; 2991 goto out; 2992 } 2993 2994 map.m_lblk = pg_start; 2995 map.m_len = pg_end - pg_start; 2996 total = 0; 2997 2998 while (map.m_lblk < pg_end) { 2999 pgoff_t idx; 3000 int cnt = 0; 3001 3002 do_map: 3003 map.m_len = pg_end - map.m_lblk; 3004 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT); 3005 if (err) 3006 goto clear_out; 3007 3008 if (!(map.m_flags & F2FS_MAP_FLAGS)) { 3009 map.m_lblk = next_pgofs; 3010 goto check; 3011 } 3012 3013 set_inode_flag(inode, FI_SKIP_WRITES); 3014 3015 idx = map.m_lblk; 3016 while (idx < map.m_lblk + map.m_len && 3017 cnt < BLKS_PER_SEG(sbi)) { 3018 struct folio *folio; 3019 3020 folio = f2fs_get_lock_data_folio(inode, idx, true); 3021 if (IS_ERR(folio)) { 3022 err = PTR_ERR(folio); 3023 goto clear_out; 3024 } 3025 3026 f2fs_folio_wait_writeback(folio, DATA, true, true); 3027 3028 folio_mark_dirty(folio); 3029 folio_set_f2fs_gcing(folio); 3030 f2fs_folio_put(folio, true); 3031 3032 idx++; 3033 cnt++; 3034 total++; 3035 } 3036 3037 map.m_lblk = idx; 3038 check: 3039 if (map.m_lblk < pg_end && cnt < BLKS_PER_SEG(sbi)) 3040 goto do_map; 3041 3042 clear_inode_flag(inode, FI_SKIP_WRITES); 3043 3044 err = filemap_fdatawrite(inode->i_mapping); 3045 if (err) 3046 goto out; 3047 } 3048 clear_out: 3049 clear_inode_flag(inode, FI_SKIP_WRITES); 3050 out: 3051 clear_inode_flag(inode, FI_OPU_WRITE); 3052 unlock_out: 3053 inode_unlock(inode); 3054 if (!err) { 3055 range->len = (u64)total << PAGE_SHIFT; 3056 stat_inc_defrag_blk_count(sbi, total); 3057 } 3058 return err; 3059 } 3060 3061 static int f2fs_ioc_defragment(struct file *filp, unsigned long arg) 3062 { 3063 struct inode *inode = file_inode(filp); 3064 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3065 struct f2fs_defragment range; 3066 int err; 3067 3068 if (!capable(CAP_SYS_ADMIN)) 3069 return -EPERM; 3070 3071 if (!S_ISREG(inode->i_mode)) 3072 return -EINVAL; 3073 3074 if (f2fs_readonly(sbi->sb)) 3075 return -EROFS; 3076 3077 if (copy_from_user(&range, (struct f2fs_defragment __user *)arg, 3078 sizeof(range))) 3079 return -EFAULT; 3080 3081 /* verify alignment of offset & size */ 3082 if (range.start & (F2FS_BLKSIZE - 1) || range.len & (F2FS_BLKSIZE - 1)) 3083 return -EINVAL; 3084 3085 if (unlikely((range.start + range.len) >> PAGE_SHIFT > 3086 max_file_blocks(inode))) 3087 return -EINVAL; 3088 3089 err = mnt_want_write_file(filp); 3090 if (err) 3091 return err; 3092 3093 err = f2fs_defragment_range(sbi, filp, &range); 3094 mnt_drop_write_file(filp); 3095 3096 if (range.len) 3097 f2fs_update_time(sbi, REQ_TIME); 3098 if (err < 0) 3099 return err; 3100 3101 if (copy_to_user((struct f2fs_defragment __user *)arg, &range, 3102 sizeof(range))) 3103 return -EFAULT; 3104 3105 return 0; 3106 } 3107 3108 static int f2fs_move_file_range(struct file *file_in, loff_t pos_in, 3109 struct file *file_out, loff_t pos_out, size_t len) 3110 { 3111 struct inode *src = file_inode(file_in); 3112 struct inode *dst = file_inode(file_out); 3113 struct f2fs_sb_info *sbi = F2FS_I_SB(src); 3114 struct f2fs_lock_context lc; 3115 size_t olen = len, dst_max_i_size = 0; 3116 size_t dst_osize; 3117 int ret; 3118 3119 if (file_in->f_path.mnt != file_out->f_path.mnt || 3120 src->i_sb != dst->i_sb) 3121 return -EXDEV; 3122 3123 if (unlikely(f2fs_readonly(src->i_sb))) 3124 return -EROFS; 3125 3126 if (!S_ISREG(src->i_mode) || !S_ISREG(dst->i_mode)) 3127 return -EINVAL; 3128 3129 if (IS_ENCRYPTED(src) || IS_ENCRYPTED(dst)) 3130 return -EOPNOTSUPP; 3131 3132 if (pos_out < 0 || pos_in < 0) 3133 return -EINVAL; 3134 3135 if (src == dst) { 3136 if (pos_in == pos_out) 3137 return 0; 3138 if (pos_out > pos_in && pos_out < pos_in + len) 3139 return -EINVAL; 3140 } 3141 3142 inode_lock(src); 3143 if (src != dst) { 3144 ret = -EBUSY; 3145 if (!inode_trylock(dst)) 3146 goto out; 3147 } 3148 3149 if (f2fs_compressed_file(src) || f2fs_compressed_file(dst) || 3150 f2fs_is_pinned_file(src) || f2fs_is_pinned_file(dst)) { 3151 ret = -EOPNOTSUPP; 3152 goto out_unlock; 3153 } 3154 3155 if (f2fs_is_atomic_file(src) || f2fs_is_atomic_file(dst)) { 3156 ret = -EINVAL; 3157 goto out_unlock; 3158 } 3159 3160 ret = -EINVAL; 3161 if (pos_in + len > src->i_size || pos_in + len < pos_in) 3162 goto out_unlock; 3163 if (len == 0) 3164 olen = len = src->i_size - pos_in; 3165 if (pos_in + len == src->i_size) 3166 len = ALIGN(src->i_size, F2FS_BLKSIZE) - pos_in; 3167 if (len == 0) { 3168 ret = 0; 3169 goto out_unlock; 3170 } 3171 3172 dst_osize = dst->i_size; 3173 if (pos_out + olen > dst->i_size) 3174 dst_max_i_size = pos_out + olen; 3175 3176 /* verify the end result is block aligned */ 3177 if (!IS_ALIGNED(pos_in, F2FS_BLKSIZE) || 3178 !IS_ALIGNED(pos_in + len, F2FS_BLKSIZE) || 3179 !IS_ALIGNED(pos_out, F2FS_BLKSIZE)) 3180 goto out_unlock; 3181 3182 ret = f2fs_convert_inline_inode(src); 3183 if (ret) 3184 goto out_unlock; 3185 3186 ret = f2fs_convert_inline_inode(dst); 3187 if (ret) 3188 goto out_unlock; 3189 3190 /* write out all dirty pages from offset */ 3191 ret = filemap_write_and_wait_range(src->i_mapping, 3192 pos_in, pos_in + len); 3193 if (ret) 3194 goto out_unlock; 3195 3196 ret = filemap_write_and_wait_range(dst->i_mapping, 3197 pos_out, pos_out + len); 3198 if (ret) 3199 goto out_unlock; 3200 3201 f2fs_balance_fs(sbi, true); 3202 3203 f2fs_down_write(&F2FS_I(src)->i_gc_rwsem[WRITE]); 3204 if (src != dst) { 3205 ret = -EBUSY; 3206 if (!f2fs_down_write_trylock(&F2FS_I(dst)->i_gc_rwsem[WRITE])) 3207 goto out_src; 3208 } 3209 3210 f2fs_lock_op(sbi, &lc); 3211 ret = __exchange_data_block(src, dst, F2FS_BYTES_TO_BLK(pos_in), 3212 F2FS_BYTES_TO_BLK(pos_out), 3213 F2FS_BYTES_TO_BLK(len), false); 3214 3215 if (!ret) { 3216 if (dst_max_i_size) 3217 f2fs_i_size_write(dst, dst_max_i_size); 3218 else if (dst_osize != dst->i_size) 3219 f2fs_i_size_write(dst, dst_osize); 3220 } 3221 f2fs_unlock_op(sbi, &lc); 3222 3223 if (src != dst) 3224 f2fs_up_write(&F2FS_I(dst)->i_gc_rwsem[WRITE]); 3225 out_src: 3226 f2fs_up_write(&F2FS_I(src)->i_gc_rwsem[WRITE]); 3227 if (ret) 3228 goto out_unlock; 3229 3230 inode_set_mtime_to_ts(src, inode_set_ctime_current(src)); 3231 f2fs_mark_inode_dirty_sync(src, false); 3232 if (src != dst) { 3233 inode_set_mtime_to_ts(dst, inode_set_ctime_current(dst)); 3234 f2fs_mark_inode_dirty_sync(dst, false); 3235 } 3236 f2fs_update_time(sbi, REQ_TIME); 3237 3238 out_unlock: 3239 if (src != dst) 3240 inode_unlock(dst); 3241 out: 3242 inode_unlock(src); 3243 return ret; 3244 } 3245 3246 static int __f2fs_ioc_move_range(struct file *filp, 3247 struct f2fs_move_range *range) 3248 { 3249 int err; 3250 3251 if (!(filp->f_mode & FMODE_READ) || 3252 !(filp->f_mode & FMODE_WRITE)) 3253 return -EBADF; 3254 3255 CLASS(fd, dst)(range->dst_fd); 3256 if (fd_empty(dst)) 3257 return -EBADF; 3258 3259 if (!(fd_file(dst)->f_mode & FMODE_WRITE)) 3260 return -EBADF; 3261 3262 err = mnt_want_write_file(filp); 3263 if (err) 3264 return err; 3265 3266 err = f2fs_move_file_range(filp, range->pos_in, fd_file(dst), 3267 range->pos_out, range->len); 3268 3269 mnt_drop_write_file(filp); 3270 return err; 3271 } 3272 3273 static int f2fs_ioc_move_range(struct file *filp, unsigned long arg) 3274 { 3275 struct f2fs_move_range range; 3276 3277 if (copy_from_user(&range, (struct f2fs_move_range __user *)arg, 3278 sizeof(range))) 3279 return -EFAULT; 3280 return __f2fs_ioc_move_range(filp, &range); 3281 } 3282 3283 static int f2fs_ioc_flush_device(struct file *filp, unsigned long arg) 3284 { 3285 struct inode *inode = file_inode(filp); 3286 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3287 struct sit_info *sm = SIT_I(sbi); 3288 unsigned int start_segno = 0, end_segno = 0; 3289 unsigned int dev_start_segno = 0, dev_end_segno = 0; 3290 struct f2fs_flush_device range; 3291 struct f2fs_gc_control gc_control = { 3292 .init_gc_type = FG_GC, 3293 .should_migrate_blocks = true, 3294 .err_gc_skipped = true, 3295 .nr_free_secs = 0 }; 3296 int ret; 3297 3298 if (!capable(CAP_SYS_ADMIN)) 3299 return -EPERM; 3300 3301 if (f2fs_readonly(sbi->sb)) 3302 return -EROFS; 3303 3304 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) 3305 return -EINVAL; 3306 3307 if (copy_from_user(&range, (struct f2fs_flush_device __user *)arg, 3308 sizeof(range))) 3309 return -EFAULT; 3310 3311 if (!f2fs_is_multi_device(sbi) || sbi->s_ndevs - 1 <= range.dev_num || 3312 __is_large_section(sbi)) { 3313 f2fs_warn(sbi, "Can't flush %u in %d for SEGS_PER_SEC %u != 1", 3314 range.dev_num, sbi->s_ndevs, SEGS_PER_SEC(sbi)); 3315 return -EINVAL; 3316 } 3317 3318 ret = mnt_want_write_file(filp); 3319 if (ret) 3320 return ret; 3321 3322 if (range.dev_num != 0) 3323 dev_start_segno = GET_SEGNO(sbi, FDEV(range.dev_num).start_blk); 3324 dev_end_segno = GET_SEGNO(sbi, FDEV(range.dev_num).end_blk); 3325 3326 start_segno = sm->last_victim[FLUSH_DEVICE]; 3327 if (start_segno < dev_start_segno || start_segno >= dev_end_segno) 3328 start_segno = dev_start_segno; 3329 end_segno = min(start_segno + range.segments, dev_end_segno); 3330 3331 while (start_segno < end_segno) { 3332 if (!f2fs_down_write_trylock_trace(&sbi->gc_lock, &gc_control.lc)) { 3333 ret = -EBUSY; 3334 goto out; 3335 } 3336 sm->last_victim[GC_CB] = end_segno + 1; 3337 sm->last_victim[GC_GREEDY] = end_segno + 1; 3338 sm->last_victim[ALLOC_NEXT] = end_segno + 1; 3339 3340 gc_control.victim_segno = start_segno; 3341 stat_inc_gc_call_count(sbi, FOREGROUND); 3342 ret = f2fs_gc(sbi, &gc_control); 3343 if (ret == -EAGAIN) 3344 ret = 0; 3345 else if (ret < 0) 3346 break; 3347 start_segno++; 3348 } 3349 out: 3350 mnt_drop_write_file(filp); 3351 return ret; 3352 } 3353 3354 static int f2fs_ioc_get_features(struct file *filp, unsigned long arg) 3355 { 3356 struct inode *inode = file_inode(filp); 3357 u32 sb_feature = le32_to_cpu(F2FS_I_SB(inode)->raw_super->feature); 3358 3359 /* Must validate to set it with SQLite behavior in Android. */ 3360 sb_feature |= F2FS_FEATURE_ATOMIC_WRITE; 3361 3362 return put_user(sb_feature, (u32 __user *)arg); 3363 } 3364 3365 #ifdef CONFIG_QUOTA 3366 int f2fs_transfer_project_quota(struct inode *inode, kprojid_t kprojid) 3367 { 3368 struct dquot *transfer_to[MAXQUOTAS] = {}; 3369 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3370 struct super_block *sb = sbi->sb; 3371 int err; 3372 3373 transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid)); 3374 if (IS_ERR(transfer_to[PRJQUOTA])) 3375 return PTR_ERR(transfer_to[PRJQUOTA]); 3376 3377 err = __dquot_transfer(inode, transfer_to); 3378 if (err) 3379 set_sbi_flag(sbi, SBI_QUOTA_NEED_REPAIR); 3380 dqput(transfer_to[PRJQUOTA]); 3381 return err; 3382 } 3383 3384 static int f2fs_ioc_setproject(struct inode *inode, __u32 projid) 3385 { 3386 struct f2fs_inode_info *fi = F2FS_I(inode); 3387 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3388 struct f2fs_inode *ri = NULL; 3389 struct f2fs_lock_context lc; 3390 kprojid_t kprojid; 3391 int err; 3392 3393 if (!f2fs_sb_has_project_quota(sbi)) { 3394 if (projid != F2FS_DEF_PROJID) 3395 return -EOPNOTSUPP; 3396 else 3397 return 0; 3398 } 3399 3400 if (!f2fs_has_extra_attr(inode)) 3401 return -EOPNOTSUPP; 3402 3403 kprojid = make_kprojid(&init_user_ns, (projid_t)projid); 3404 3405 if (projid_eq(kprojid, fi->i_projid)) 3406 return 0; 3407 3408 err = -EPERM; 3409 /* Is it quota file? Do not allow user to mess with it */ 3410 if (IS_NOQUOTA(inode)) 3411 return err; 3412 3413 if (!F2FS_FITS_IN_INODE(ri, fi->i_extra_isize, i_projid)) 3414 return -EOVERFLOW; 3415 3416 err = f2fs_dquot_initialize(inode); 3417 if (err) 3418 return err; 3419 3420 f2fs_lock_op(sbi, &lc); 3421 err = f2fs_transfer_project_quota(inode, kprojid); 3422 if (err) 3423 goto out_unlock; 3424 3425 fi->i_projid = kprojid; 3426 inode_set_ctime_current(inode); 3427 f2fs_mark_inode_dirty_sync(inode, true); 3428 out_unlock: 3429 f2fs_unlock_op(sbi, &lc); 3430 return err; 3431 } 3432 #else 3433 int f2fs_transfer_project_quota(struct inode *inode, kprojid_t kprojid) 3434 { 3435 return 0; 3436 } 3437 3438 static int f2fs_ioc_setproject(struct inode *inode, __u32 projid) 3439 { 3440 if (projid != F2FS_DEF_PROJID) 3441 return -EOPNOTSUPP; 3442 return 0; 3443 } 3444 #endif 3445 3446 int f2fs_fileattr_get(struct dentry *dentry, struct file_kattr *fa) 3447 { 3448 struct inode *inode = d_inode(dentry); 3449 struct f2fs_inode_info *fi = F2FS_I(inode); 3450 u32 fsflags = f2fs_iflags_to_fsflags(fi->i_flags); 3451 3452 if (IS_ENCRYPTED(inode)) 3453 fsflags |= FS_ENCRYPT_FL; 3454 if (IS_VERITY(inode)) 3455 fsflags |= FS_VERITY_FL; 3456 if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) 3457 fsflags |= FS_INLINE_DATA_FL; 3458 if (is_inode_flag_set(inode, FI_PIN_FILE)) 3459 fsflags |= FS_NOCOW_FL; 3460 3461 fileattr_fill_flags(fa, fsflags & F2FS_GETTABLE_FS_FL); 3462 3463 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode))) 3464 fa->fsx_projid = from_kprojid(&init_user_ns, fi->i_projid); 3465 3466 return 0; 3467 } 3468 3469 int f2fs_fileattr_set(struct mnt_idmap *idmap, 3470 struct dentry *dentry, struct file_kattr *fa) 3471 { 3472 struct inode *inode = d_inode(dentry); 3473 u32 fsflags = fa->flags, mask = F2FS_SETTABLE_FS_FL; 3474 u32 iflags; 3475 int err; 3476 3477 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) 3478 return -EIO; 3479 if (!f2fs_is_checkpoint_ready(F2FS_I_SB(inode))) 3480 return -ENOSPC; 3481 if (fsflags & ~F2FS_GETTABLE_FS_FL) 3482 return -EOPNOTSUPP; 3483 fsflags &= F2FS_SETTABLE_FS_FL; 3484 if (!fa->flags_valid) 3485 mask &= FS_COMMON_FL; 3486 3487 iflags = f2fs_fsflags_to_iflags(fsflags); 3488 if (f2fs_mask_flags(inode->i_mode, iflags) != iflags) 3489 return -EOPNOTSUPP; 3490 3491 err = f2fs_setflags_common(inode, iflags, f2fs_fsflags_to_iflags(mask)); 3492 if (!err) 3493 err = f2fs_ioc_setproject(inode, fa->fsx_projid); 3494 3495 return err; 3496 } 3497 3498 int f2fs_pin_file_control(struct inode *inode, bool inc) 3499 { 3500 struct f2fs_inode_info *fi = F2FS_I(inode); 3501 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3502 3503 if (IS_DEVICE_ALIASING(inode)) 3504 return -EINVAL; 3505 3506 if (fi->i_gc_failures >= sbi->gc_pin_file_threshold) { 3507 f2fs_warn(sbi, "%s: Enable GC = ino %llx after %x GC trials", 3508 __func__, inode->i_ino, fi->i_gc_failures); 3509 clear_inode_flag(inode, FI_PIN_FILE); 3510 return -EAGAIN; 3511 } 3512 3513 /* Use i_gc_failures for normal file as a risk signal. */ 3514 if (inc) 3515 f2fs_i_gc_failures_write(inode, fi->i_gc_failures + 1); 3516 3517 return 0; 3518 } 3519 3520 static int f2fs_ioc_set_pin_file(struct file *filp, unsigned long arg) 3521 { 3522 struct inode *inode = file_inode(filp); 3523 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3524 __u32 pin; 3525 int ret = 0; 3526 3527 if (get_user(pin, (__u32 __user *)arg)) 3528 return -EFAULT; 3529 3530 if (!S_ISREG(inode->i_mode)) 3531 return -EINVAL; 3532 3533 if (f2fs_readonly(sbi->sb)) 3534 return -EROFS; 3535 3536 if (!pin && IS_DEVICE_ALIASING(inode)) 3537 return -EOPNOTSUPP; 3538 3539 ret = mnt_want_write_file(filp); 3540 if (ret) 3541 return ret; 3542 3543 inode_lock(inode); 3544 3545 if (f2fs_is_atomic_file(inode)) { 3546 ret = -EINVAL; 3547 goto out; 3548 } 3549 3550 if (!pin) { 3551 clear_inode_flag(inode, FI_PIN_FILE); 3552 f2fs_i_gc_failures_write(inode, 0); 3553 goto done; 3554 } else if (f2fs_is_pinned_file(inode)) { 3555 goto done; 3556 } 3557 3558 if (F2FS_HAS_BLOCKS(inode)) { 3559 ret = -EFBIG; 3560 goto out; 3561 } 3562 3563 /* Let's allow file pinning on zoned device. */ 3564 if (!f2fs_sb_has_blkzoned(sbi) && 3565 f2fs_should_update_outplace(inode, NULL)) { 3566 ret = -EINVAL; 3567 goto out; 3568 } 3569 3570 if (f2fs_pin_file_control(inode, false)) { 3571 ret = -EAGAIN; 3572 goto out; 3573 } 3574 3575 ret = f2fs_convert_inline_inode(inode); 3576 if (ret) 3577 goto out; 3578 3579 if (!f2fs_disable_compressed_file(inode)) { 3580 ret = -EOPNOTSUPP; 3581 goto out; 3582 } 3583 3584 set_inode_flag(inode, FI_PIN_FILE); 3585 ret = F2FS_I(inode)->i_gc_failures; 3586 done: 3587 f2fs_update_time(sbi, REQ_TIME); 3588 out: 3589 inode_unlock(inode); 3590 mnt_drop_write_file(filp); 3591 return ret; 3592 } 3593 3594 static int f2fs_ioc_get_pin_file(struct file *filp, unsigned long arg) 3595 { 3596 struct inode *inode = file_inode(filp); 3597 __u32 pin = 0; 3598 3599 if (is_inode_flag_set(inode, FI_PIN_FILE)) 3600 pin = F2FS_I(inode)->i_gc_failures; 3601 return put_user(pin, (u32 __user *)arg); 3602 } 3603 3604 static int f2fs_ioc_get_dev_alias_file(struct file *filp, unsigned long arg) 3605 { 3606 return put_user(IS_DEVICE_ALIASING(file_inode(filp)) ? 1 : 0, 3607 (u32 __user *)arg); 3608 } 3609 3610 static int f2fs_ioc_io_prio(struct file *filp, unsigned long arg) 3611 { 3612 struct inode *inode = file_inode(filp); 3613 __u32 level; 3614 3615 if (get_user(level, (__u32 __user *)arg)) 3616 return -EFAULT; 3617 3618 if (!S_ISREG(inode->i_mode) || level >= F2FS_IOPRIO_MAX) 3619 return -EINVAL; 3620 3621 inode_lock(inode); 3622 F2FS_I(inode)->ioprio_hint = level; 3623 inode_unlock(inode); 3624 return 0; 3625 } 3626 3627 int f2fs_precache_extents(struct inode *inode) 3628 { 3629 struct f2fs_inode_info *fi = F2FS_I(inode); 3630 struct f2fs_map_blocks map; 3631 pgoff_t m_next_extent; 3632 loff_t end; 3633 int err; 3634 3635 if (is_inode_flag_set(inode, FI_NO_EXTENT)) 3636 return -EOPNOTSUPP; 3637 3638 map.m_lblk = 0; 3639 map.m_pblk = 0; 3640 map.m_next_pgofs = NULL; 3641 map.m_next_extent = &m_next_extent; 3642 map.m_seg_type = NO_CHECK_TYPE; 3643 map.m_may_create = false; 3644 end = F2FS_BLK_ALIGN(i_size_read(inode)); 3645 3646 while (map.m_lblk < end) { 3647 map.m_len = end - map.m_lblk; 3648 3649 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 3650 err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_PRECACHE); 3651 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 3652 if (err || !map.m_len) 3653 return err; 3654 3655 map.m_lblk = m_next_extent; 3656 } 3657 3658 return 0; 3659 } 3660 3661 static int f2fs_ioc_precache_extents(struct file *filp) 3662 { 3663 return f2fs_precache_extents(file_inode(filp)); 3664 } 3665 3666 static int f2fs_ioc_resize_fs(struct file *filp, unsigned long arg) 3667 { 3668 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp)); 3669 __u64 block_count; 3670 3671 if (!capable(CAP_SYS_ADMIN)) 3672 return -EPERM; 3673 3674 if (f2fs_readonly(sbi->sb)) 3675 return -EROFS; 3676 3677 if (copy_from_user(&block_count, (void __user *)arg, 3678 sizeof(block_count))) 3679 return -EFAULT; 3680 3681 return f2fs_resize_fs(filp, block_count); 3682 } 3683 3684 static int f2fs_ioc_enable_verity(struct file *filp, unsigned long arg) 3685 { 3686 struct inode *inode = file_inode(filp); 3687 3688 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME); 3689 3690 if (!f2fs_sb_has_verity(F2FS_I_SB(inode))) { 3691 f2fs_warn(F2FS_I_SB(inode), 3692 "Can't enable fs-verity on inode %llu: the verity feature is not enabled on this filesystem", 3693 inode->i_ino); 3694 return -EOPNOTSUPP; 3695 } 3696 3697 return fsverity_ioctl_enable(filp, (const void __user *)arg); 3698 } 3699 3700 static int f2fs_ioc_measure_verity(struct file *filp, unsigned long arg) 3701 { 3702 if (!f2fs_sb_has_verity(F2FS_I_SB(file_inode(filp)))) 3703 return -EOPNOTSUPP; 3704 3705 return fsverity_ioctl_measure(filp, (void __user *)arg); 3706 } 3707 3708 static int f2fs_ioc_read_verity_metadata(struct file *filp, unsigned long arg) 3709 { 3710 if (!f2fs_sb_has_verity(F2FS_I_SB(file_inode(filp)))) 3711 return -EOPNOTSUPP; 3712 3713 return fsverity_ioctl_read_metadata(filp, (const void __user *)arg); 3714 } 3715 3716 static int f2fs_ioc_getfslabel(struct file *filp, unsigned long arg) 3717 { 3718 struct inode *inode = file_inode(filp); 3719 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3720 char *vbuf; 3721 int count; 3722 int err = 0; 3723 3724 vbuf = f2fs_kzalloc(sbi, MAX_VOLUME_NAME, GFP_KERNEL); 3725 if (!vbuf) 3726 return -ENOMEM; 3727 3728 f2fs_down_read(&sbi->sb_lock); 3729 count = utf16s_to_utf8s(sbi->raw_super->volume_name, 3730 ARRAY_SIZE(sbi->raw_super->volume_name), 3731 UTF16_LITTLE_ENDIAN, vbuf, MAX_VOLUME_NAME); 3732 f2fs_up_read(&sbi->sb_lock); 3733 3734 if (copy_to_user((char __user *)arg, vbuf, 3735 min(FSLABEL_MAX, count))) 3736 err = -EFAULT; 3737 3738 kfree(vbuf); 3739 return err; 3740 } 3741 3742 static int f2fs_ioc_setfslabel(struct file *filp, unsigned long arg) 3743 { 3744 struct inode *inode = file_inode(filp); 3745 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3746 char *vbuf; 3747 int err = 0; 3748 3749 if (!capable(CAP_SYS_ADMIN)) 3750 return -EPERM; 3751 3752 vbuf = strndup_user((const char __user *)arg, FSLABEL_MAX); 3753 if (IS_ERR(vbuf)) 3754 return PTR_ERR(vbuf); 3755 3756 err = mnt_want_write_file(filp); 3757 if (err) 3758 goto out; 3759 3760 f2fs_down_write(&sbi->sb_lock); 3761 3762 memset(sbi->raw_super->volume_name, 0, 3763 sizeof(sbi->raw_super->volume_name)); 3764 utf8s_to_utf16s(vbuf, strlen(vbuf), UTF16_LITTLE_ENDIAN, 3765 sbi->raw_super->volume_name, 3766 ARRAY_SIZE(sbi->raw_super->volume_name)); 3767 3768 err = f2fs_commit_super(sbi, false); 3769 3770 f2fs_up_write(&sbi->sb_lock); 3771 3772 mnt_drop_write_file(filp); 3773 out: 3774 kfree(vbuf); 3775 return err; 3776 } 3777 3778 static int f2fs_get_compress_blocks(struct inode *inode, __u64 *blocks) 3779 { 3780 if (!f2fs_sb_has_compression(F2FS_I_SB(inode))) 3781 return -EOPNOTSUPP; 3782 3783 if (!f2fs_compressed_file(inode)) 3784 return -EINVAL; 3785 3786 *blocks = atomic_read(&F2FS_I(inode)->i_compr_blocks); 3787 3788 return 0; 3789 } 3790 3791 static int f2fs_ioc_get_compress_blocks(struct file *filp, unsigned long arg) 3792 { 3793 struct inode *inode = file_inode(filp); 3794 __u64 blocks; 3795 int ret; 3796 3797 ret = f2fs_get_compress_blocks(inode, &blocks); 3798 if (ret < 0) 3799 return ret; 3800 3801 return put_user(blocks, (u64 __user *)arg); 3802 } 3803 3804 static int release_compress_blocks(struct dnode_of_data *dn, pgoff_t count) 3805 { 3806 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 3807 unsigned int released_blocks = 0; 3808 int cluster_size = F2FS_I(dn->inode)->i_cluster_size; 3809 block_t blkaddr; 3810 int i; 3811 3812 for (i = 0; i < count; i++) { 3813 blkaddr = data_blkaddr(dn->inode, dn->node_folio, 3814 dn->ofs_in_node + i); 3815 3816 if (!__is_valid_data_blkaddr(blkaddr)) 3817 continue; 3818 if (unlikely(!f2fs_is_valid_blkaddr(sbi, blkaddr, 3819 DATA_GENERIC_ENHANCE))) 3820 return -EFSCORRUPTED; 3821 } 3822 3823 while (count) { 3824 int compr_blocks = 0; 3825 3826 for (i = 0; i < cluster_size; i++, dn->ofs_in_node++) { 3827 blkaddr = f2fs_data_blkaddr(dn); 3828 3829 if (i == 0) { 3830 if (blkaddr == COMPRESS_ADDR) 3831 continue; 3832 dn->ofs_in_node += cluster_size; 3833 goto next; 3834 } 3835 3836 if (__is_valid_data_blkaddr(blkaddr)) 3837 compr_blocks++; 3838 3839 if (blkaddr != NEW_ADDR) 3840 continue; 3841 3842 f2fs_set_data_blkaddr(dn, NULL_ADDR); 3843 } 3844 3845 f2fs_i_compr_blocks_update(dn->inode, compr_blocks, false); 3846 dec_valid_block_count(sbi, dn->inode, 3847 cluster_size - compr_blocks); 3848 3849 released_blocks += cluster_size - compr_blocks; 3850 next: 3851 count -= cluster_size; 3852 } 3853 3854 return released_blocks; 3855 } 3856 3857 static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) 3858 { 3859 struct inode *inode = file_inode(filp); 3860 struct f2fs_inode_info *fi = F2FS_I(inode); 3861 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 3862 struct f2fs_lock_context lc; 3863 pgoff_t page_idx = 0, last_idx; 3864 unsigned int released_blocks = 0; 3865 int ret; 3866 int writecount; 3867 3868 if (!f2fs_sb_has_compression(sbi)) 3869 return -EOPNOTSUPP; 3870 3871 if (f2fs_readonly(sbi->sb)) 3872 return -EROFS; 3873 3874 ret = mnt_want_write_file(filp); 3875 if (ret) 3876 return ret; 3877 3878 f2fs_balance_fs(sbi, true); 3879 3880 inode_lock(inode); 3881 3882 writecount = atomic_read(&inode->i_writecount); 3883 if ((filp->f_mode & FMODE_WRITE && writecount != 1) || 3884 (!(filp->f_mode & FMODE_WRITE) && writecount)) { 3885 ret = -EBUSY; 3886 goto out; 3887 } 3888 3889 if (!f2fs_compressed_file(inode) || 3890 is_inode_flag_set(inode, FI_COMPRESS_RELEASED)) { 3891 ret = -EINVAL; 3892 goto out; 3893 } 3894 3895 ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 3896 if (ret) 3897 goto out; 3898 3899 if (!atomic_read(&fi->i_compr_blocks)) { 3900 ret = -EPERM; 3901 goto out; 3902 } 3903 3904 set_inode_flag(inode, FI_COMPRESS_RELEASED); 3905 inode_set_ctime_current(inode); 3906 f2fs_mark_inode_dirty_sync(inode, true); 3907 3908 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 3909 filemap_invalidate_lock(inode->i_mapping); 3910 3911 last_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 3912 3913 while (page_idx < last_idx) { 3914 struct dnode_of_data dn; 3915 pgoff_t end_offset, count; 3916 3917 f2fs_lock_op(sbi, &lc); 3918 3919 set_new_dnode(&dn, inode, NULL, NULL, 0); 3920 ret = f2fs_get_dnode_of_data(&dn, page_idx, LOOKUP_NODE); 3921 if (ret) { 3922 f2fs_unlock_op(sbi, &lc); 3923 if (ret == -ENOENT) { 3924 page_idx = f2fs_get_next_page_offset(&dn, 3925 page_idx); 3926 ret = 0; 3927 continue; 3928 } 3929 break; 3930 } 3931 3932 end_offset = ADDRS_PER_PAGE(dn.node_folio, inode); 3933 count = min(end_offset - dn.ofs_in_node, last_idx - page_idx); 3934 count = round_up(count, fi->i_cluster_size); 3935 3936 ret = release_compress_blocks(&dn, count); 3937 3938 f2fs_put_dnode(&dn); 3939 3940 f2fs_unlock_op(sbi, &lc); 3941 3942 if (ret < 0) 3943 break; 3944 3945 page_idx += count; 3946 released_blocks += ret; 3947 } 3948 3949 filemap_invalidate_unlock(inode->i_mapping); 3950 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 3951 out: 3952 if (released_blocks) 3953 f2fs_update_time(sbi, REQ_TIME); 3954 inode_unlock(inode); 3955 3956 mnt_drop_write_file(filp); 3957 3958 if (ret >= 0) { 3959 ret = put_user(released_blocks, (u64 __user *)arg); 3960 } else if (released_blocks && 3961 atomic_read(&fi->i_compr_blocks)) { 3962 set_sbi_flag(sbi, SBI_NEED_FSCK); 3963 f2fs_warn(sbi, "%s: partial blocks were released i_ino=%llx " 3964 "iblocks=%llu, released=%u, compr_blocks=%u, " 3965 "run fsck to fix.", 3966 __func__, inode->i_ino, inode->i_blocks, 3967 released_blocks, 3968 atomic_read(&fi->i_compr_blocks)); 3969 } 3970 3971 return ret; 3972 } 3973 3974 static int reserve_compress_blocks(struct dnode_of_data *dn, pgoff_t count, 3975 unsigned int *reserved_blocks) 3976 { 3977 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 3978 int cluster_size = F2FS_I(dn->inode)->i_cluster_size; 3979 block_t blkaddr; 3980 int i; 3981 3982 for (i = 0; i < count; i++) { 3983 blkaddr = data_blkaddr(dn->inode, dn->node_folio, 3984 dn->ofs_in_node + i); 3985 3986 if (!__is_valid_data_blkaddr(blkaddr)) 3987 continue; 3988 if (unlikely(!f2fs_is_valid_blkaddr(sbi, blkaddr, 3989 DATA_GENERIC_ENHANCE))) 3990 return -EFSCORRUPTED; 3991 } 3992 3993 while (count) { 3994 int compr_blocks = 0; 3995 blkcnt_t reserved = 0; 3996 blkcnt_t to_reserved; 3997 int ret; 3998 3999 for (i = 0; i < cluster_size; i++) { 4000 blkaddr = data_blkaddr(dn->inode, dn->node_folio, 4001 dn->ofs_in_node + i); 4002 4003 if (i == 0) { 4004 if (blkaddr != COMPRESS_ADDR) { 4005 dn->ofs_in_node += cluster_size; 4006 goto next; 4007 } 4008 continue; 4009 } 4010 4011 /* 4012 * compressed cluster was not released due to it 4013 * fails in release_compress_blocks(), so NEW_ADDR 4014 * is a possible case. 4015 */ 4016 if (blkaddr == NEW_ADDR) { 4017 reserved++; 4018 continue; 4019 } 4020 if (__is_valid_data_blkaddr(blkaddr)) { 4021 compr_blocks++; 4022 continue; 4023 } 4024 } 4025 4026 to_reserved = cluster_size - compr_blocks - reserved; 4027 4028 /* for the case all blocks in cluster were reserved */ 4029 if (reserved && to_reserved == 1) { 4030 dn->ofs_in_node += cluster_size; 4031 goto next; 4032 } 4033 4034 ret = inc_valid_block_count(sbi, dn->inode, 4035 &to_reserved, false); 4036 if (unlikely(ret)) 4037 return ret; 4038 4039 for (i = 0; i < cluster_size; i++, dn->ofs_in_node++) { 4040 if (f2fs_data_blkaddr(dn) == NULL_ADDR) 4041 f2fs_set_data_blkaddr(dn, NEW_ADDR); 4042 } 4043 4044 f2fs_i_compr_blocks_update(dn->inode, compr_blocks, true); 4045 4046 *reserved_blocks += to_reserved; 4047 next: 4048 count -= cluster_size; 4049 } 4050 4051 return 0; 4052 } 4053 4054 static int f2fs_reserve_compress_blocks(struct file *filp, unsigned long arg) 4055 { 4056 struct inode *inode = file_inode(filp); 4057 struct f2fs_inode_info *fi = F2FS_I(inode); 4058 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4059 pgoff_t page_idx = 0, last_idx; 4060 unsigned int reserved_blocks = 0; 4061 int ret; 4062 4063 if (!f2fs_sb_has_compression(sbi)) 4064 return -EOPNOTSUPP; 4065 4066 if (f2fs_readonly(sbi->sb)) 4067 return -EROFS; 4068 4069 ret = mnt_want_write_file(filp); 4070 if (ret) 4071 return ret; 4072 4073 f2fs_balance_fs(sbi, true); 4074 4075 inode_lock(inode); 4076 4077 if (!f2fs_compressed_file(inode) || 4078 !is_inode_flag_set(inode, FI_COMPRESS_RELEASED)) { 4079 ret = -EINVAL; 4080 goto unlock_inode; 4081 } 4082 4083 if (atomic_read(&fi->i_compr_blocks)) 4084 goto unlock_inode; 4085 4086 f2fs_down_write(&fi->i_gc_rwsem[WRITE]); 4087 filemap_invalidate_lock(inode->i_mapping); 4088 4089 last_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 4090 4091 while (page_idx < last_idx) { 4092 struct dnode_of_data dn; 4093 struct f2fs_lock_context lc; 4094 pgoff_t end_offset, count; 4095 4096 f2fs_lock_op(sbi, &lc); 4097 4098 set_new_dnode(&dn, inode, NULL, NULL, 0); 4099 ret = f2fs_get_dnode_of_data(&dn, page_idx, LOOKUP_NODE); 4100 if (ret) { 4101 f2fs_unlock_op(sbi, &lc); 4102 if (ret == -ENOENT) { 4103 page_idx = f2fs_get_next_page_offset(&dn, 4104 page_idx); 4105 ret = 0; 4106 continue; 4107 } 4108 break; 4109 } 4110 4111 end_offset = ADDRS_PER_PAGE(dn.node_folio, inode); 4112 count = min(end_offset - dn.ofs_in_node, last_idx - page_idx); 4113 count = round_up(count, fi->i_cluster_size); 4114 4115 ret = reserve_compress_blocks(&dn, count, &reserved_blocks); 4116 4117 f2fs_put_dnode(&dn); 4118 4119 f2fs_unlock_op(sbi, &lc); 4120 4121 if (ret < 0) 4122 break; 4123 4124 page_idx += count; 4125 } 4126 4127 filemap_invalidate_unlock(inode->i_mapping); 4128 f2fs_up_write(&fi->i_gc_rwsem[WRITE]); 4129 4130 if (!ret) { 4131 clear_inode_flag(inode, FI_COMPRESS_RELEASED); 4132 inode_set_ctime_current(inode); 4133 f2fs_mark_inode_dirty_sync(inode, true); 4134 } 4135 unlock_inode: 4136 if (reserved_blocks) 4137 f2fs_update_time(sbi, REQ_TIME); 4138 inode_unlock(inode); 4139 mnt_drop_write_file(filp); 4140 4141 if (!ret) { 4142 ret = put_user(reserved_blocks, (u64 __user *)arg); 4143 } else if (reserved_blocks && 4144 atomic_read(&fi->i_compr_blocks)) { 4145 set_sbi_flag(sbi, SBI_NEED_FSCK); 4146 f2fs_warn(sbi, "%s: partial blocks were reserved i_ino=%llx " 4147 "iblocks=%llu, reserved=%u, compr_blocks=%u, " 4148 "run fsck to fix.", 4149 __func__, inode->i_ino, inode->i_blocks, 4150 reserved_blocks, 4151 atomic_read(&fi->i_compr_blocks)); 4152 } 4153 4154 return ret; 4155 } 4156 4157 static int f2fs_secure_erase(struct block_device *bdev, struct inode *inode, 4158 pgoff_t off, block_t block, block_t len, u32 flags) 4159 { 4160 sector_t sector = SECTOR_FROM_BLOCK(block); 4161 sector_t nr_sects = SECTOR_FROM_BLOCK(len); 4162 int ret = 0; 4163 4164 if (flags & F2FS_TRIM_FILE_DISCARD) { 4165 if (bdev_max_secure_erase_sectors(bdev)) 4166 ret = blkdev_issue_secure_erase(bdev, sector, nr_sects, 4167 GFP_NOFS); 4168 else 4169 ret = blkdev_issue_discard(bdev, sector, nr_sects, 4170 GFP_NOFS); 4171 } 4172 4173 if (!ret && (flags & F2FS_TRIM_FILE_ZEROOUT)) { 4174 if (IS_ENCRYPTED(inode)) 4175 ret = fscrypt_zeroout_range(inode, 4176 (loff_t)off << inode->i_blkbits, sector, 4177 (u64)len << inode->i_blkbits); 4178 else 4179 ret = blkdev_issue_zeroout(bdev, sector, nr_sects, 4180 GFP_NOFS, 0); 4181 } 4182 4183 return ret; 4184 } 4185 4186 static int f2fs_sec_trim_file(struct file *filp, unsigned long arg) 4187 { 4188 struct inode *inode = file_inode(filp); 4189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4190 struct address_space *mapping = inode->i_mapping; 4191 struct block_device *prev_bdev = NULL; 4192 struct f2fs_sectrim_range range; 4193 pgoff_t index, pg_end, prev_index = 0; 4194 block_t prev_block = 0, len = 0; 4195 loff_t end_addr; 4196 bool to_end = false; 4197 int ret = 0; 4198 4199 if (!(filp->f_mode & FMODE_WRITE)) 4200 return -EBADF; 4201 4202 if (copy_from_user(&range, (struct f2fs_sectrim_range __user *)arg, 4203 sizeof(range))) 4204 return -EFAULT; 4205 4206 if (range.flags == 0 || (range.flags & ~F2FS_TRIM_FILE_MASK) || 4207 !S_ISREG(inode->i_mode)) 4208 return -EINVAL; 4209 4210 if (((range.flags & F2FS_TRIM_FILE_DISCARD) && 4211 !f2fs_hw_support_discard(sbi)) || 4212 ((range.flags & F2FS_TRIM_FILE_ZEROOUT) && 4213 IS_ENCRYPTED(inode) && f2fs_is_multi_device(sbi))) 4214 return -EOPNOTSUPP; 4215 4216 ret = mnt_want_write_file(filp); 4217 if (ret) 4218 return ret; 4219 inode_lock(inode); 4220 4221 if (f2fs_is_atomic_file(inode) || f2fs_compressed_file(inode) || 4222 range.start >= inode->i_size) { 4223 ret = -EINVAL; 4224 goto err; 4225 } 4226 4227 if (range.len == 0) 4228 goto err; 4229 4230 if (inode->i_size - range.start > range.len) { 4231 end_addr = range.start + range.len; 4232 } else { 4233 end_addr = range.len == (u64)-1 ? 4234 sbi->sb->s_maxbytes : inode->i_size; 4235 to_end = true; 4236 } 4237 4238 if (!IS_ALIGNED(range.start, F2FS_BLKSIZE) || 4239 (!to_end && !IS_ALIGNED(end_addr, F2FS_BLKSIZE))) { 4240 ret = -EINVAL; 4241 goto err; 4242 } 4243 4244 index = F2FS_BYTES_TO_BLK(range.start); 4245 pg_end = DIV_ROUND_UP(end_addr, F2FS_BLKSIZE); 4246 4247 ret = f2fs_convert_inline_inode(inode); 4248 if (ret) 4249 goto err; 4250 4251 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 4252 filemap_invalidate_lock(mapping); 4253 4254 ret = filemap_write_and_wait_range(mapping, range.start, 4255 to_end ? LLONG_MAX : end_addr - 1); 4256 if (ret) 4257 goto out; 4258 4259 truncate_inode_pages_range(mapping, range.start, 4260 to_end ? -1 : end_addr - 1); 4261 4262 while (index < pg_end) { 4263 struct dnode_of_data dn; 4264 pgoff_t end_offset, count; 4265 int i; 4266 4267 set_new_dnode(&dn, inode, NULL, NULL, 0); 4268 ret = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE); 4269 if (ret) { 4270 if (ret == -ENOENT) { 4271 index = f2fs_get_next_page_offset(&dn, index); 4272 continue; 4273 } 4274 goto out; 4275 } 4276 4277 end_offset = ADDRS_PER_PAGE(dn.node_folio, inode); 4278 count = min(end_offset - dn.ofs_in_node, pg_end - index); 4279 for (i = 0; i < count; i++, index++, dn.ofs_in_node++) { 4280 struct block_device *cur_bdev; 4281 block_t blkaddr = f2fs_data_blkaddr(&dn); 4282 4283 if (!__is_valid_data_blkaddr(blkaddr)) 4284 continue; 4285 4286 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, 4287 DATA_GENERIC_ENHANCE)) { 4288 ret = -EFSCORRUPTED; 4289 f2fs_put_dnode(&dn); 4290 goto out; 4291 } 4292 4293 cur_bdev = f2fs_target_device(sbi, blkaddr, NULL); 4294 if (f2fs_is_multi_device(sbi)) { 4295 int di = f2fs_target_device_index(sbi, blkaddr); 4296 4297 blkaddr -= FDEV(di).start_blk; 4298 } 4299 4300 if (len) { 4301 if (prev_bdev == cur_bdev && 4302 index == prev_index + len && 4303 blkaddr == prev_block + len) { 4304 len++; 4305 } else { 4306 ret = f2fs_secure_erase(prev_bdev, 4307 inode, prev_index, prev_block, 4308 len, range.flags); 4309 if (ret) { 4310 f2fs_put_dnode(&dn); 4311 goto out; 4312 } 4313 4314 len = 0; 4315 } 4316 } 4317 4318 if (!len) { 4319 prev_bdev = cur_bdev; 4320 prev_index = index; 4321 prev_block = blkaddr; 4322 len = 1; 4323 } 4324 } 4325 4326 f2fs_put_dnode(&dn); 4327 4328 if (fatal_signal_pending(current)) { 4329 ret = -EINTR; 4330 goto out; 4331 } 4332 cond_resched(); 4333 } 4334 4335 if (len) 4336 ret = f2fs_secure_erase(prev_bdev, inode, prev_index, 4337 prev_block, len, range.flags); 4338 f2fs_update_time(sbi, REQ_TIME); 4339 out: 4340 filemap_invalidate_unlock(mapping); 4341 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 4342 err: 4343 inode_unlock(inode); 4344 mnt_drop_write_file(filp); 4345 4346 return ret; 4347 } 4348 4349 static int f2fs_ioc_get_compress_option(struct file *filp, unsigned long arg) 4350 { 4351 struct inode *inode = file_inode(filp); 4352 struct f2fs_comp_option option; 4353 4354 if (!f2fs_sb_has_compression(F2FS_I_SB(inode))) 4355 return -EOPNOTSUPP; 4356 4357 inode_lock_shared(inode); 4358 4359 if (!f2fs_compressed_file(inode)) { 4360 inode_unlock_shared(inode); 4361 return -ENODATA; 4362 } 4363 4364 option.algorithm = F2FS_I(inode)->i_compress_algorithm; 4365 option.log_cluster_size = F2FS_I(inode)->i_log_cluster_size; 4366 4367 inode_unlock_shared(inode); 4368 4369 if (copy_to_user((struct f2fs_comp_option __user *)arg, &option, 4370 sizeof(option))) 4371 return -EFAULT; 4372 4373 return 0; 4374 } 4375 4376 static int f2fs_ioc_set_compress_option(struct file *filp, unsigned long arg) 4377 { 4378 struct inode *inode = file_inode(filp); 4379 struct f2fs_inode_info *fi = F2FS_I(inode); 4380 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4381 struct f2fs_comp_option option; 4382 int ret = 0; 4383 4384 if (!f2fs_sb_has_compression(sbi)) 4385 return -EOPNOTSUPP; 4386 4387 if (!(filp->f_mode & FMODE_WRITE)) 4388 return -EBADF; 4389 4390 if (copy_from_user(&option, (struct f2fs_comp_option __user *)arg, 4391 sizeof(option))) 4392 return -EFAULT; 4393 4394 if (option.log_cluster_size < MIN_COMPRESS_LOG_SIZE || 4395 option.log_cluster_size > MAX_COMPRESS_LOG_SIZE || 4396 option.algorithm >= COMPRESS_MAX) 4397 return -EINVAL; 4398 4399 ret = mnt_want_write_file(filp); 4400 if (ret) 4401 return ret; 4402 inode_lock(inode); 4403 4404 f2fs_down_write(&F2FS_I(inode)->i_sem); 4405 if (!f2fs_compressed_file(inode)) { 4406 ret = -EINVAL; 4407 goto out; 4408 } 4409 4410 if (f2fs_is_mmap_file(inode) || get_dirty_pages(inode)) { 4411 ret = -EBUSY; 4412 goto out; 4413 } 4414 4415 if (F2FS_HAS_BLOCKS(inode)) { 4416 ret = -EFBIG; 4417 goto out; 4418 } 4419 4420 fi->i_compress_algorithm = option.algorithm; 4421 fi->i_log_cluster_size = option.log_cluster_size; 4422 fi->i_cluster_size = BIT(option.log_cluster_size); 4423 /* Set default level */ 4424 if (fi->i_compress_algorithm == COMPRESS_ZSTD) 4425 fi->i_compress_level = F2FS_ZSTD_DEFAULT_CLEVEL; 4426 else 4427 fi->i_compress_level = 0; 4428 /* Adjust mount option level */ 4429 if (option.algorithm == F2FS_OPTION(sbi).compress_algorithm && 4430 F2FS_OPTION(sbi).compress_level) 4431 fi->i_compress_level = F2FS_OPTION(sbi).compress_level; 4432 f2fs_mark_inode_dirty_sync(inode, true); 4433 4434 if (!f2fs_is_compress_backend_ready(inode)) 4435 f2fs_warn(sbi, "compression algorithm is successfully set, " 4436 "but current kernel doesn't support this algorithm."); 4437 out: 4438 f2fs_up_write(&fi->i_sem); 4439 inode_unlock(inode); 4440 mnt_drop_write_file(filp); 4441 4442 return ret; 4443 } 4444 4445 static int redirty_blocks(struct inode *inode, pgoff_t page_idx, int len) 4446 { 4447 DEFINE_READAHEAD(ractl, NULL, NULL, inode->i_mapping, page_idx); 4448 struct address_space *mapping = inode->i_mapping; 4449 struct folio *folio; 4450 pgoff_t redirty_idx = page_idx; 4451 int page_len = 0, ret = 0; 4452 4453 filemap_invalidate_lock_shared(mapping); 4454 page_cache_ra_unbounded(&ractl, len, 0); 4455 filemap_invalidate_unlock_shared(mapping); 4456 4457 do { 4458 folio = read_cache_folio(mapping, page_idx, NULL, NULL); 4459 if (IS_ERR(folio)) { 4460 ret = PTR_ERR(folio); 4461 break; 4462 } 4463 page_len += folio_nr_pages(folio) - (page_idx - folio->index); 4464 page_idx = folio_next_index(folio); 4465 } while (page_len < len); 4466 4467 do { 4468 folio = filemap_lock_folio(mapping, redirty_idx); 4469 4470 /* It will never fail, when folio has pinned above */ 4471 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(folio)); 4472 4473 f2fs_folio_wait_writeback(folio, DATA, true, true); 4474 4475 folio_mark_dirty(folio); 4476 folio_set_f2fs_gcing(folio); 4477 redirty_idx = folio_next_index(folio); 4478 folio_unlock(folio); 4479 folio_put_refs(folio, 2); 4480 } while (redirty_idx < page_idx); 4481 4482 return ret; 4483 } 4484 4485 static int f2fs_ioc_decompress_file(struct file *filp) 4486 { 4487 struct inode *inode = file_inode(filp); 4488 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4489 struct f2fs_inode_info *fi = F2FS_I(inode); 4490 pgoff_t page_idx = 0, last_idx, cluster_idx; 4491 int ret; 4492 4493 if (!f2fs_sb_has_compression(sbi) || 4494 F2FS_OPTION(sbi).compress_mode != COMPR_MODE_USER) 4495 return -EOPNOTSUPP; 4496 4497 if (!(filp->f_mode & FMODE_WRITE)) 4498 return -EBADF; 4499 4500 f2fs_balance_fs(sbi, true); 4501 4502 ret = mnt_want_write_file(filp); 4503 if (ret) 4504 return ret; 4505 inode_lock(inode); 4506 4507 if (!f2fs_is_compress_backend_ready(inode)) { 4508 ret = -EOPNOTSUPP; 4509 goto out; 4510 } 4511 4512 if (!f2fs_compressed_file(inode) || 4513 is_inode_flag_set(inode, FI_COMPRESS_RELEASED)) { 4514 ret = -EINVAL; 4515 goto out; 4516 } 4517 4518 ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 4519 if (ret) 4520 goto out; 4521 4522 if (!atomic_read(&fi->i_compr_blocks)) 4523 goto out; 4524 4525 last_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 4526 last_idx >>= fi->i_log_cluster_size; 4527 4528 for (cluster_idx = 0; cluster_idx < last_idx; cluster_idx++) { 4529 page_idx = cluster_idx << fi->i_log_cluster_size; 4530 4531 if (!f2fs_is_compressed_cluster(inode, page_idx)) 4532 continue; 4533 4534 ret = redirty_blocks(inode, page_idx, fi->i_cluster_size); 4535 if (ret < 0) 4536 break; 4537 4538 if (get_dirty_pages(inode) >= BLKS_PER_SEG(sbi)) { 4539 ret = filemap_fdatawrite(inode->i_mapping); 4540 if (ret < 0) 4541 break; 4542 } 4543 4544 cond_resched(); 4545 if (fatal_signal_pending(current)) { 4546 ret = -EINTR; 4547 break; 4548 } 4549 } 4550 4551 if (!ret) 4552 ret = filemap_write_and_wait_range(inode->i_mapping, 0, 4553 LLONG_MAX); 4554 4555 if (ret) 4556 f2fs_warn(sbi, "%s: The file might be partially decompressed (errno=%d). Please delete the file.", 4557 __func__, ret); 4558 f2fs_update_time(sbi, REQ_TIME); 4559 out: 4560 inode_unlock(inode); 4561 mnt_drop_write_file(filp); 4562 4563 return ret; 4564 } 4565 4566 static int f2fs_ioc_compress_file(struct file *filp) 4567 { 4568 struct inode *inode = file_inode(filp); 4569 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4570 struct f2fs_inode_info *fi = F2FS_I(inode); 4571 pgoff_t page_idx = 0, last_idx, cluster_idx; 4572 int ret; 4573 4574 if (!f2fs_sb_has_compression(sbi) || 4575 F2FS_OPTION(sbi).compress_mode != COMPR_MODE_USER) 4576 return -EOPNOTSUPP; 4577 4578 if (!(filp->f_mode & FMODE_WRITE)) 4579 return -EBADF; 4580 4581 f2fs_balance_fs(sbi, true); 4582 4583 ret = mnt_want_write_file(filp); 4584 if (ret) 4585 return ret; 4586 inode_lock(inode); 4587 4588 if (!f2fs_is_compress_backend_ready(inode)) { 4589 ret = -EOPNOTSUPP; 4590 goto out; 4591 } 4592 4593 if (!f2fs_compressed_file(inode) || 4594 is_inode_flag_set(inode, FI_COMPRESS_RELEASED)) { 4595 ret = -EINVAL; 4596 goto out; 4597 } 4598 4599 ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX); 4600 if (ret) 4601 goto out; 4602 4603 set_inode_flag(inode, FI_ENABLE_COMPRESS); 4604 4605 last_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 4606 last_idx >>= fi->i_log_cluster_size; 4607 4608 for (cluster_idx = 0; cluster_idx < last_idx; cluster_idx++) { 4609 page_idx = cluster_idx << fi->i_log_cluster_size; 4610 4611 if (f2fs_is_sparse_cluster(inode, page_idx)) 4612 continue; 4613 4614 ret = redirty_blocks(inode, page_idx, fi->i_cluster_size); 4615 if (ret < 0) 4616 break; 4617 4618 if (get_dirty_pages(inode) >= BLKS_PER_SEG(sbi)) { 4619 ret = filemap_fdatawrite(inode->i_mapping); 4620 if (ret < 0) 4621 break; 4622 } 4623 4624 cond_resched(); 4625 if (fatal_signal_pending(current)) { 4626 ret = -EINTR; 4627 break; 4628 } 4629 } 4630 4631 if (!ret) 4632 ret = filemap_write_and_wait_range(inode->i_mapping, 0, 4633 LLONG_MAX); 4634 4635 clear_inode_flag(inode, FI_ENABLE_COMPRESS); 4636 4637 if (ret) 4638 f2fs_warn(sbi, "%s: The file might be partially compressed (errno=%d). Please delete the file.", 4639 __func__, ret); 4640 f2fs_update_time(sbi, REQ_TIME); 4641 out: 4642 inode_unlock(inode); 4643 mnt_drop_write_file(filp); 4644 4645 return ret; 4646 } 4647 4648 static long __f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 4649 { 4650 switch (cmd) { 4651 case FS_IOC_GETVERSION: 4652 return f2fs_ioc_getversion(filp, arg); 4653 case F2FS_IOC_START_ATOMIC_WRITE: 4654 return f2fs_ioc_start_atomic_write(filp, false); 4655 case F2FS_IOC_START_ATOMIC_REPLACE: 4656 return f2fs_ioc_start_atomic_write(filp, true); 4657 case F2FS_IOC_COMMIT_ATOMIC_WRITE: 4658 return f2fs_ioc_commit_atomic_write(filp); 4659 case F2FS_IOC_ABORT_ATOMIC_WRITE: 4660 return f2fs_ioc_abort_atomic_write(filp); 4661 case F2FS_IOC_START_VOLATILE_WRITE: 4662 case F2FS_IOC_RELEASE_VOLATILE_WRITE: 4663 return -EOPNOTSUPP; 4664 case F2FS_IOC_SHUTDOWN: 4665 return f2fs_ioc_shutdown(filp, arg); 4666 case FITRIM: 4667 return f2fs_ioc_fitrim(filp, arg); 4668 case FS_IOC_SET_ENCRYPTION_POLICY: 4669 return f2fs_ioc_set_encryption_policy(filp, arg); 4670 case FS_IOC_GET_ENCRYPTION_POLICY: 4671 return f2fs_ioc_get_encryption_policy(filp, arg); 4672 case FS_IOC_GET_ENCRYPTION_PWSALT: 4673 return f2fs_ioc_get_encryption_pwsalt(filp, arg); 4674 case FS_IOC_GET_ENCRYPTION_POLICY_EX: 4675 return f2fs_ioc_get_encryption_policy_ex(filp, arg); 4676 case FS_IOC_ADD_ENCRYPTION_KEY: 4677 return f2fs_ioc_add_encryption_key(filp, arg); 4678 case FS_IOC_REMOVE_ENCRYPTION_KEY: 4679 return f2fs_ioc_remove_encryption_key(filp, arg); 4680 case FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS: 4681 return f2fs_ioc_remove_encryption_key_all_users(filp, arg); 4682 case FS_IOC_GET_ENCRYPTION_KEY_STATUS: 4683 return f2fs_ioc_get_encryption_key_status(filp, arg); 4684 case FS_IOC_GET_ENCRYPTION_NONCE: 4685 return f2fs_ioc_get_encryption_nonce(filp, arg); 4686 case F2FS_IOC_GARBAGE_COLLECT: 4687 return f2fs_ioc_gc(filp, arg); 4688 case F2FS_IOC_GARBAGE_COLLECT_RANGE: 4689 return f2fs_ioc_gc_range(filp, arg); 4690 case F2FS_IOC_WRITE_CHECKPOINT: 4691 return f2fs_ioc_write_checkpoint(filp); 4692 case F2FS_IOC_DEFRAGMENT: 4693 return f2fs_ioc_defragment(filp, arg); 4694 case F2FS_IOC_MOVE_RANGE: 4695 return f2fs_ioc_move_range(filp, arg); 4696 case F2FS_IOC_FLUSH_DEVICE: 4697 return f2fs_ioc_flush_device(filp, arg); 4698 case F2FS_IOC_GET_FEATURES: 4699 return f2fs_ioc_get_features(filp, arg); 4700 case F2FS_IOC_GET_PIN_FILE: 4701 return f2fs_ioc_get_pin_file(filp, arg); 4702 case F2FS_IOC_SET_PIN_FILE: 4703 return f2fs_ioc_set_pin_file(filp, arg); 4704 case F2FS_IOC_PRECACHE_EXTENTS: 4705 return f2fs_ioc_precache_extents(filp); 4706 case F2FS_IOC_RESIZE_FS: 4707 return f2fs_ioc_resize_fs(filp, arg); 4708 case FS_IOC_ENABLE_VERITY: 4709 return f2fs_ioc_enable_verity(filp, arg); 4710 case FS_IOC_MEASURE_VERITY: 4711 return f2fs_ioc_measure_verity(filp, arg); 4712 case FS_IOC_READ_VERITY_METADATA: 4713 return f2fs_ioc_read_verity_metadata(filp, arg); 4714 case FS_IOC_GETFSLABEL: 4715 return f2fs_ioc_getfslabel(filp, arg); 4716 case FS_IOC_SETFSLABEL: 4717 return f2fs_ioc_setfslabel(filp, arg); 4718 case F2FS_IOC_GET_COMPRESS_BLOCKS: 4719 return f2fs_ioc_get_compress_blocks(filp, arg); 4720 case F2FS_IOC_RELEASE_COMPRESS_BLOCKS: 4721 return f2fs_release_compress_blocks(filp, arg); 4722 case F2FS_IOC_RESERVE_COMPRESS_BLOCKS: 4723 return f2fs_reserve_compress_blocks(filp, arg); 4724 case F2FS_IOC_SEC_TRIM_FILE: 4725 return f2fs_sec_trim_file(filp, arg); 4726 case F2FS_IOC_GET_COMPRESS_OPTION: 4727 return f2fs_ioc_get_compress_option(filp, arg); 4728 case F2FS_IOC_SET_COMPRESS_OPTION: 4729 return f2fs_ioc_set_compress_option(filp, arg); 4730 case F2FS_IOC_DECOMPRESS_FILE: 4731 return f2fs_ioc_decompress_file(filp); 4732 case F2FS_IOC_COMPRESS_FILE: 4733 return f2fs_ioc_compress_file(filp); 4734 case F2FS_IOC_GET_DEV_ALIAS_FILE: 4735 return f2fs_ioc_get_dev_alias_file(filp, arg); 4736 case F2FS_IOC_IO_PRIO: 4737 return f2fs_ioc_io_prio(filp, arg); 4738 default: 4739 return -ENOTTY; 4740 } 4741 } 4742 4743 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 4744 { 4745 if (unlikely(f2fs_cp_error(F2FS_I_SB(file_inode(filp))))) 4746 return -EIO; 4747 if (!f2fs_is_checkpoint_ready(F2FS_I_SB(file_inode(filp)))) 4748 return -ENOSPC; 4749 4750 return __f2fs_ioctl(filp, cmd, arg); 4751 } 4752 4753 /* 4754 * Return %true if the given read or write request should use direct I/O, or 4755 * %false if it should use buffered I/O. 4756 */ 4757 static bool f2fs_should_use_dio(struct inode *inode, struct kiocb *iocb, 4758 struct iov_iter *iter) 4759 { 4760 unsigned int align; 4761 4762 if (!(iocb->ki_flags & IOCB_DIRECT)) 4763 return false; 4764 4765 if (f2fs_force_buffered_io(inode, iov_iter_rw(iter))) 4766 return false; 4767 4768 /* 4769 * Direct I/O not aligned to the disk's logical_block_size will be 4770 * attempted, but will fail with -EINVAL. 4771 * 4772 * f2fs additionally requires that direct I/O be aligned to the 4773 * filesystem block size, which is often a stricter requirement. 4774 * However, f2fs traditionally falls back to buffered I/O on requests 4775 * that are logical_block_size-aligned but not fs-block aligned. 4776 * 4777 * The below logic implements this behavior. 4778 */ 4779 align = iocb->ki_pos | iov_iter_alignment(iter); 4780 if (!IS_ALIGNED(align, i_blocksize(inode)) && 4781 IS_ALIGNED(align, bdev_logical_block_size(inode->i_sb->s_bdev))) 4782 return false; 4783 4784 return true; 4785 } 4786 4787 static int f2fs_dio_read_end_io(struct kiocb *iocb, ssize_t size, int error, 4788 unsigned int flags) 4789 { 4790 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(iocb->ki_filp)); 4791 4792 dec_page_count(sbi, F2FS_DIO_READ); 4793 if (error) 4794 return error; 4795 f2fs_update_iostat(sbi, NULL, APP_DIRECT_READ_IO, size); 4796 return 0; 4797 } 4798 4799 static const struct iomap_dio_ops f2fs_iomap_dio_read_ops = { 4800 .end_io = f2fs_dio_read_end_io, 4801 }; 4802 4803 static ssize_t f2fs_dio_read_iter(struct kiocb *iocb, struct iov_iter *to) 4804 { 4805 struct file *file = iocb->ki_filp; 4806 struct inode *inode = file_inode(file); 4807 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4808 struct f2fs_inode_info *fi = F2FS_I(inode); 4809 const loff_t pos = iocb->ki_pos; 4810 const size_t count = iov_iter_count(to); 4811 struct iomap_dio *dio; 4812 ssize_t ret; 4813 4814 if (count == 0) 4815 return 0; /* skip atime update */ 4816 4817 trace_f2fs_direct_IO_enter(inode, iocb, count, READ); 4818 4819 if (iocb->ki_flags & IOCB_NOWAIT) { 4820 if (!f2fs_down_read_trylock(&fi->i_gc_rwsem[READ])) { 4821 ret = -EAGAIN; 4822 goto out; 4823 } 4824 } else { 4825 f2fs_down_read(&fi->i_gc_rwsem[READ]); 4826 } 4827 4828 /* dio is not compatible w/ atomic file */ 4829 if (f2fs_is_atomic_file(inode)) { 4830 f2fs_up_read(&fi->i_gc_rwsem[READ]); 4831 ret = -EOPNOTSUPP; 4832 goto out; 4833 } 4834 4835 /* 4836 * We have to use __iomap_dio_rw() and iomap_dio_complete() instead of 4837 * the higher-level function iomap_dio_rw() in order to ensure that the 4838 * F2FS_DIO_READ counter will be decremented correctly in all cases. 4839 */ 4840 inc_page_count(sbi, F2FS_DIO_READ); 4841 dio = __iomap_dio_rw(iocb, to, &f2fs_iomap_ops, 4842 &f2fs_iomap_dio_read_ops, 0, NULL, 0); 4843 if (IS_ERR_OR_NULL(dio)) { 4844 ret = PTR_ERR_OR_ZERO(dio); 4845 if (ret != -EIOCBQUEUED) 4846 dec_page_count(sbi, F2FS_DIO_READ); 4847 } else { 4848 ret = iomap_dio_complete(dio); 4849 } 4850 4851 f2fs_up_read(&fi->i_gc_rwsem[READ]); 4852 4853 file_accessed(file); 4854 out: 4855 trace_f2fs_direct_IO_exit(inode, pos, count, READ, ret); 4856 return ret; 4857 } 4858 4859 static void f2fs_trace_rw_file_path(struct file *file, loff_t pos, size_t count, 4860 int rw) 4861 { 4862 struct inode *inode = file_inode(file); 4863 char *buf, *path; 4864 4865 buf = f2fs_getname(F2FS_I_SB(inode)); 4866 if (!buf) 4867 return; 4868 path = dentry_path_raw(file_dentry(file), buf, PATH_MAX); 4869 if (IS_ERR(path)) 4870 goto free_buf; 4871 if (rw == WRITE) 4872 trace_f2fs_datawrite_start(inode, pos, count, 4873 current->pid, path, current->comm); 4874 else 4875 trace_f2fs_dataread_start(inode, pos, count, 4876 current->pid, path, current->comm); 4877 free_buf: 4878 f2fs_putname(buf); 4879 } 4880 4881 static ssize_t f2fs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) 4882 { 4883 struct inode *inode = file_inode(iocb->ki_filp); 4884 const loff_t pos = iocb->ki_pos; 4885 ssize_t ret; 4886 bool dio; 4887 4888 if (!f2fs_is_compress_backend_ready(inode)) 4889 return -EOPNOTSUPP; 4890 4891 if (trace_f2fs_dataread_start_enabled()) 4892 f2fs_trace_rw_file_path(iocb->ki_filp, iocb->ki_pos, 4893 iov_iter_count(to), READ); 4894 4895 dio = f2fs_should_use_dio(inode, iocb, to); 4896 4897 /* In LFS mode, if there is inflight dio, wait for its completion */ 4898 if (f2fs_lfs_mode(F2FS_I_SB(inode)) && 4899 get_pages(F2FS_I_SB(inode), F2FS_DIO_WRITE) && 4900 (!f2fs_is_pinned_file(inode) || !dio)) 4901 inode_dio_wait(inode); 4902 4903 if (dio) { 4904 ret = f2fs_dio_read_iter(iocb, to); 4905 } else { 4906 ret = filemap_read(iocb, to, 0); 4907 if (ret > 0) 4908 f2fs_update_iostat(F2FS_I_SB(inode), inode, 4909 APP_BUFFERED_READ_IO, ret); 4910 } 4911 trace_f2fs_dataread_end(inode, pos, ret); 4912 return ret; 4913 } 4914 4915 static ssize_t f2fs_file_splice_read(struct file *in, loff_t *ppos, 4916 struct pipe_inode_info *pipe, 4917 size_t len, unsigned int flags) 4918 { 4919 struct inode *inode = file_inode(in); 4920 const loff_t pos = *ppos; 4921 ssize_t ret; 4922 4923 if (!f2fs_is_compress_backend_ready(inode)) 4924 return -EOPNOTSUPP; 4925 4926 if (trace_f2fs_dataread_start_enabled()) 4927 f2fs_trace_rw_file_path(in, pos, len, READ); 4928 4929 ret = filemap_splice_read(in, ppos, pipe, len, flags); 4930 if (ret > 0) 4931 f2fs_update_iostat(F2FS_I_SB(inode), inode, 4932 APP_BUFFERED_READ_IO, ret); 4933 4934 trace_f2fs_dataread_end(inode, pos, ret); 4935 return ret; 4936 } 4937 4938 static ssize_t f2fs_write_checks(struct kiocb *iocb, struct iov_iter *from) 4939 { 4940 struct file *file = iocb->ki_filp; 4941 struct inode *inode = file_inode(file); 4942 ssize_t count; 4943 int err; 4944 4945 if (IS_IMMUTABLE(inode)) 4946 return -EPERM; 4947 4948 if (is_inode_flag_set(inode, FI_COMPRESS_RELEASED)) 4949 return -EPERM; 4950 4951 count = generic_write_checks(iocb, from); 4952 if (count <= 0) 4953 return count; 4954 4955 err = file_modified(file); 4956 if (err) 4957 return err; 4958 4959 f2fs_zero_post_eof_page(inode, 4960 iocb->ki_pos + iov_iter_count(from), true); 4961 return count; 4962 } 4963 4964 /* 4965 * Preallocate blocks for a write request, if it is possible and helpful to do 4966 * so. Returns a positive number if blocks may have been preallocated, 0 if no 4967 * blocks were preallocated, or a negative errno value if something went 4968 * seriously wrong. Also sets FI_PREALLOCATED_ALL on the inode if *all* the 4969 * requested blocks (not just some of them) have been allocated. 4970 */ 4971 static int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *iter, 4972 bool dio) 4973 { 4974 struct inode *inode = file_inode(iocb->ki_filp); 4975 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 4976 const loff_t pos = iocb->ki_pos; 4977 const size_t count = iov_iter_count(iter); 4978 struct f2fs_map_blocks map = {}; 4979 int flag; 4980 int ret; 4981 4982 /* If it will be an out-of-place direct write, don't bother. */ 4983 if (dio && f2fs_lfs_mode(sbi)) 4984 return 0; 4985 /* 4986 * Don't preallocate holes aligned to DIO_SKIP_HOLES which turns into 4987 * buffered IO, if DIO meets any holes. 4988 */ 4989 if (dio && i_size_read(inode) && 4990 (F2FS_BYTES_TO_BLK(pos) < F2FS_BLK_ALIGN(i_size_read(inode)))) 4991 return 0; 4992 4993 /* No-wait I/O can't allocate blocks. */ 4994 if (iocb->ki_flags & IOCB_NOWAIT) 4995 return 0; 4996 4997 /* If it will be a short write, don't bother. */ 4998 if (fault_in_iov_iter_readable(iter, count)) 4999 return 0; 5000 5001 if (f2fs_has_inline_data(inode)) { 5002 /* If the data will fit inline, don't bother. */ 5003 if (pos + count <= MAX_INLINE_DATA(inode)) 5004 return 0; 5005 ret = f2fs_convert_inline_inode(inode); 5006 if (ret) 5007 return ret; 5008 } 5009 5010 /* Do not preallocate blocks that will be written partially in 4KB. */ 5011 map.m_lblk = F2FS_BLK_ALIGN(pos); 5012 map.m_len = F2FS_BYTES_TO_BLK(pos + count); 5013 if (map.m_len > map.m_lblk) 5014 map.m_len -= map.m_lblk; 5015 else 5016 return 0; 5017 5018 if (!IS_DEVICE_ALIASING(inode)) 5019 map.m_may_create = true; 5020 if (dio) { 5021 map.m_seg_type = f2fs_rw_hint_to_seg_type(sbi, 5022 inode->i_write_hint); 5023 flag = F2FS_GET_BLOCK_PRE_DIO; 5024 } else { 5025 map.m_seg_type = NO_CHECK_TYPE; 5026 flag = F2FS_GET_BLOCK_PRE_AIO; 5027 } 5028 5029 ret = f2fs_map_blocks(inode, &map, flag); 5030 /* -ENOSPC|-EDQUOT are fine to report the number of allocated blocks. */ 5031 if (ret < 0 && !((ret == -ENOSPC || ret == -EDQUOT) && map.m_len > 0)) 5032 return ret; 5033 if (ret == 0) 5034 set_inode_flag(inode, FI_PREALLOCATED_ALL); 5035 return map.m_len; 5036 } 5037 5038 static ssize_t f2fs_buffered_write_iter(struct kiocb *iocb, 5039 struct iov_iter *from) 5040 { 5041 struct file *file = iocb->ki_filp; 5042 struct inode *inode = file_inode(file); 5043 ssize_t ret; 5044 5045 if (iocb->ki_flags & IOCB_NOWAIT) 5046 return -EOPNOTSUPP; 5047 5048 ret = generic_perform_write(iocb, from); 5049 5050 if (ret > 0) { 5051 f2fs_update_iostat(F2FS_I_SB(inode), inode, 5052 APP_BUFFERED_IO, ret); 5053 } 5054 return ret; 5055 } 5056 5057 static int f2fs_dio_write_end_io(struct kiocb *iocb, ssize_t size, int error, 5058 unsigned int flags) 5059 { 5060 struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(iocb->ki_filp)); 5061 5062 dec_page_count(sbi, F2FS_DIO_WRITE); 5063 if (error) 5064 return error; 5065 f2fs_update_time(sbi, REQ_TIME); 5066 f2fs_update_iostat(sbi, NULL, APP_DIRECT_IO, size); 5067 return 0; 5068 } 5069 5070 static void f2fs_dio_write_submit_io(const struct iomap_iter *iter, 5071 struct bio *bio, loff_t file_offset) 5072 { 5073 struct inode *inode = iter->inode; 5074 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 5075 enum log_type type = f2fs_rw_hint_to_seg_type(sbi, inode->i_write_hint); 5076 enum temp_type temp = f2fs_get_segment_temp(sbi, type); 5077 5078 bio->bi_write_hint = f2fs_io_type_to_rw_hint(sbi, DATA, temp); 5079 blk_crypto_submit_bio(bio); 5080 } 5081 5082 static const struct iomap_dio_ops f2fs_iomap_dio_write_ops = { 5083 .end_io = f2fs_dio_write_end_io, 5084 .submit_io = f2fs_dio_write_submit_io, 5085 }; 5086 5087 static void f2fs_flush_buffered_write(struct address_space *mapping, 5088 loff_t start_pos, loff_t end_pos) 5089 { 5090 int ret; 5091 5092 ret = filemap_write_and_wait_range(mapping, start_pos, end_pos); 5093 if (ret < 0) 5094 return; 5095 invalidate_mapping_pages(mapping, 5096 start_pos >> PAGE_SHIFT, 5097 end_pos >> PAGE_SHIFT); 5098 } 5099 5100 static ssize_t f2fs_dio_write_iter(struct kiocb *iocb, struct iov_iter *from, 5101 bool *may_need_sync) 5102 { 5103 struct file *file = iocb->ki_filp; 5104 struct inode *inode = file_inode(file); 5105 struct f2fs_inode_info *fi = F2FS_I(inode); 5106 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 5107 const bool do_opu = f2fs_lfs_mode(sbi); 5108 const loff_t pos = iocb->ki_pos; 5109 const ssize_t count = iov_iter_count(from); 5110 unsigned int dio_flags; 5111 struct iomap_dio *dio; 5112 ssize_t ret; 5113 5114 trace_f2fs_direct_IO_enter(inode, iocb, count, WRITE); 5115 5116 if (iocb->ki_flags & IOCB_NOWAIT) { 5117 /* f2fs_convert_inline_inode() and block allocation can block */ 5118 if (f2fs_has_inline_data(inode) || 5119 !f2fs_overwrite_io(inode, pos, count)) { 5120 ret = -EAGAIN; 5121 goto out; 5122 } 5123 5124 if (!f2fs_down_read_trylock(&fi->i_gc_rwsem[WRITE])) { 5125 ret = -EAGAIN; 5126 goto out; 5127 } 5128 if (do_opu && !f2fs_down_read_trylock(&fi->i_gc_rwsem[READ])) { 5129 f2fs_up_read(&fi->i_gc_rwsem[WRITE]); 5130 ret = -EAGAIN; 5131 goto out; 5132 } 5133 } else { 5134 ret = f2fs_convert_inline_inode(inode); 5135 if (ret) 5136 goto out; 5137 5138 f2fs_down_read(&fi->i_gc_rwsem[WRITE]); 5139 if (do_opu) 5140 f2fs_down_read(&fi->i_gc_rwsem[READ]); 5141 } 5142 5143 /* 5144 * We have to use __iomap_dio_rw() and iomap_dio_complete() instead of 5145 * the higher-level function iomap_dio_rw() in order to ensure that the 5146 * F2FS_DIO_WRITE counter will be decremented correctly in all cases. 5147 */ 5148 inc_page_count(sbi, F2FS_DIO_WRITE); 5149 dio_flags = 0; 5150 if (pos + count > inode->i_size) 5151 dio_flags |= IOMAP_DIO_FORCE_WAIT; 5152 dio = __iomap_dio_rw(iocb, from, &f2fs_iomap_ops, 5153 &f2fs_iomap_dio_write_ops, dio_flags, NULL, 0); 5154 if (IS_ERR_OR_NULL(dio)) { 5155 ret = PTR_ERR_OR_ZERO(dio); 5156 if (ret == -ENOTBLK) 5157 ret = 0; 5158 if (ret != -EIOCBQUEUED) 5159 dec_page_count(sbi, F2FS_DIO_WRITE); 5160 } else { 5161 ret = iomap_dio_complete(dio); 5162 } 5163 5164 if (do_opu) 5165 f2fs_up_read(&fi->i_gc_rwsem[READ]); 5166 f2fs_up_read(&fi->i_gc_rwsem[WRITE]); 5167 5168 if (ret < 0) 5169 goto out; 5170 if (pos + ret > inode->i_size) 5171 f2fs_i_size_write(inode, pos + ret); 5172 if (!do_opu) 5173 set_inode_flag(inode, FI_UPDATE_WRITE); 5174 5175 if (iov_iter_count(from)) { 5176 ssize_t ret2; 5177 loff_t bufio_start_pos = iocb->ki_pos; 5178 5179 /* 5180 * The direct write was partial, so we need to fall back to a 5181 * buffered write for the remainder. 5182 */ 5183 5184 ret2 = f2fs_buffered_write_iter(iocb, from); 5185 if (iov_iter_count(from)) 5186 f2fs_write_failed(inode, iocb->ki_pos); 5187 if (ret2 < 0) 5188 goto out; 5189 5190 /* 5191 * Ensure that the pagecache pages are written to disk and 5192 * invalidated to preserve the expected O_DIRECT semantics. 5193 */ 5194 if (ret2 > 0) { 5195 loff_t bufio_end_pos = bufio_start_pos + ret2 - 1; 5196 5197 ret += ret2; 5198 5199 f2fs_flush_buffered_write(file->f_mapping, 5200 bufio_start_pos, 5201 bufio_end_pos); 5202 } 5203 } else { 5204 /* iomap_dio_rw() already handled the generic_write_sync(). */ 5205 *may_need_sync = false; 5206 } 5207 out: 5208 trace_f2fs_direct_IO_exit(inode, pos, count, WRITE, ret); 5209 return ret; 5210 } 5211 5212 static ssize_t f2fs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) 5213 { 5214 struct inode *inode = file_inode(iocb->ki_filp); 5215 const loff_t orig_pos = iocb->ki_pos; 5216 const size_t orig_count = iov_iter_count(from); 5217 loff_t target_size; 5218 bool dio; 5219 bool may_need_sync = true; 5220 int preallocated; 5221 const loff_t pos = iocb->ki_pos; 5222 const ssize_t count = iov_iter_count(from); 5223 ssize_t ret; 5224 5225 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) { 5226 ret = -EIO; 5227 goto out; 5228 } 5229 5230 if (!f2fs_is_compress_backend_ready(inode)) { 5231 ret = -EOPNOTSUPP; 5232 goto out; 5233 } 5234 5235 if (iocb->ki_flags & IOCB_NOWAIT) { 5236 if (!inode_trylock(inode)) { 5237 ret = -EAGAIN; 5238 goto out; 5239 } 5240 } else { 5241 inode_lock(inode); 5242 } 5243 5244 if (f2fs_is_pinned_file(inode) && 5245 !f2fs_overwrite_io(inode, pos, count)) { 5246 ret = -EIO; 5247 goto out_unlock; 5248 } 5249 5250 ret = f2fs_write_checks(iocb, from); 5251 if (ret <= 0) 5252 goto out_unlock; 5253 5254 /* Determine whether we will do a direct write or a buffered write. */ 5255 dio = f2fs_should_use_dio(inode, iocb, from); 5256 5257 /* dio is not compatible w/ atomic write */ 5258 if (dio && f2fs_is_atomic_file(inode)) { 5259 ret = -EOPNOTSUPP; 5260 goto out_unlock; 5261 } 5262 5263 /* Possibly preallocate the blocks for the write. */ 5264 target_size = iocb->ki_pos + iov_iter_count(from); 5265 preallocated = f2fs_preallocate_blocks(iocb, from, dio); 5266 if (preallocated < 0) { 5267 ret = preallocated; 5268 } else { 5269 if (trace_f2fs_datawrite_start_enabled()) 5270 f2fs_trace_rw_file_path(iocb->ki_filp, iocb->ki_pos, 5271 orig_count, WRITE); 5272 5273 /* Do the actual write. */ 5274 ret = dio ? 5275 f2fs_dio_write_iter(iocb, from, &may_need_sync) : 5276 f2fs_buffered_write_iter(iocb, from); 5277 5278 trace_f2fs_datawrite_end(inode, orig_pos, ret); 5279 } 5280 5281 /* Don't leave any preallocated blocks around past i_size. */ 5282 if (preallocated && i_size_read(inode) < target_size) { 5283 f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 5284 filemap_invalidate_lock(inode->i_mapping); 5285 if (!f2fs_truncate(inode)) 5286 file_dont_truncate(inode); 5287 filemap_invalidate_unlock(inode->i_mapping); 5288 f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); 5289 } else { 5290 file_dont_truncate(inode); 5291 } 5292 5293 clear_inode_flag(inode, FI_PREALLOCATED_ALL); 5294 out_unlock: 5295 inode_unlock(inode); 5296 out: 5297 trace_f2fs_file_write_iter(inode, orig_pos, orig_count, ret); 5298 5299 if (ret > 0 && may_need_sync) 5300 ret = generic_write_sync(iocb, ret); 5301 5302 /* If buffered IO was forced, flush and drop the data from 5303 * the page cache to preserve O_DIRECT semantics 5304 */ 5305 if (ret > 0 && !dio && (iocb->ki_flags & IOCB_DIRECT)) 5306 f2fs_flush_buffered_write(iocb->ki_filp->f_mapping, 5307 orig_pos, 5308 orig_pos + ret - 1); 5309 5310 return ret; 5311 } 5312 5313 static int f2fs_file_fadvise(struct file *filp, loff_t offset, loff_t len, 5314 int advice) 5315 { 5316 struct address_space *mapping; 5317 struct backing_dev_info *bdi; 5318 struct inode *inode = file_inode(filp); 5319 int err; 5320 5321 trace_f2fs_fadvise(inode, offset, len, advice); 5322 5323 if (advice == POSIX_FADV_SEQUENTIAL) { 5324 if (S_ISFIFO(inode->i_mode)) 5325 return -ESPIPE; 5326 5327 mapping = filp->f_mapping; 5328 if (!mapping || len < 0) 5329 return -EINVAL; 5330 5331 bdi = inode_to_bdi(mapping->host); 5332 filp->f_ra.ra_pages = bdi->ra_pages * 5333 F2FS_I_SB(inode)->seq_file_ra_mul; 5334 spin_lock(&filp->f_lock); 5335 filp->f_mode &= ~FMODE_RANDOM; 5336 spin_unlock(&filp->f_lock); 5337 return 0; 5338 } else if (advice == POSIX_FADV_WILLNEED && offset == 0) { 5339 /* Load extent cache at the first readahead. */ 5340 f2fs_precache_extents(inode); 5341 } 5342 5343 err = generic_fadvise(filp, offset, len, advice); 5344 if (err) 5345 return err; 5346 5347 if (advice == POSIX_FADV_DONTNEED && 5348 (test_opt(F2FS_I_SB(inode), COMPRESS_CACHE) && 5349 f2fs_compressed_file(inode))) 5350 f2fs_invalidate_compress_pages(F2FS_I_SB(inode), inode->i_ino); 5351 else if (advice == POSIX_FADV_NOREUSE) 5352 err = f2fs_keep_noreuse_range(inode, offset, len); 5353 return err; 5354 } 5355 5356 #ifdef CONFIG_COMPAT 5357 struct compat_f2fs_gc_range { 5358 u32 sync; 5359 compat_u64 start; 5360 compat_u64 len; 5361 }; 5362 #define F2FS_IOC32_GARBAGE_COLLECT_RANGE _IOW(F2FS_IOCTL_MAGIC, 11,\ 5363 struct compat_f2fs_gc_range) 5364 5365 static int f2fs_compat_ioc_gc_range(struct file *file, unsigned long arg) 5366 { 5367 struct compat_f2fs_gc_range __user *urange; 5368 struct f2fs_gc_range range; 5369 int err; 5370 5371 urange = compat_ptr(arg); 5372 err = get_user(range.sync, &urange->sync); 5373 err |= get_user(range.start, &urange->start); 5374 err |= get_user(range.len, &urange->len); 5375 if (err) 5376 return -EFAULT; 5377 5378 return __f2fs_ioc_gc_range(file, &range); 5379 } 5380 5381 struct compat_f2fs_move_range { 5382 u32 dst_fd; 5383 compat_u64 pos_in; 5384 compat_u64 pos_out; 5385 compat_u64 len; 5386 }; 5387 #define F2FS_IOC32_MOVE_RANGE _IOWR(F2FS_IOCTL_MAGIC, 9, \ 5388 struct compat_f2fs_move_range) 5389 5390 static int f2fs_compat_ioc_move_range(struct file *file, unsigned long arg) 5391 { 5392 struct compat_f2fs_move_range __user *urange; 5393 struct f2fs_move_range range; 5394 int err; 5395 5396 urange = compat_ptr(arg); 5397 err = get_user(range.dst_fd, &urange->dst_fd); 5398 err |= get_user(range.pos_in, &urange->pos_in); 5399 err |= get_user(range.pos_out, &urange->pos_out); 5400 err |= get_user(range.len, &urange->len); 5401 if (err) 5402 return -EFAULT; 5403 5404 return __f2fs_ioc_move_range(file, &range); 5405 } 5406 5407 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 5408 { 5409 if (unlikely(f2fs_cp_error(F2FS_I_SB(file_inode(file))))) 5410 return -EIO; 5411 if (!f2fs_is_checkpoint_ready(F2FS_I_SB(file_inode(file)))) 5412 return -ENOSPC; 5413 5414 switch (cmd) { 5415 case FS_IOC32_GETVERSION: 5416 cmd = FS_IOC_GETVERSION; 5417 break; 5418 case F2FS_IOC32_GARBAGE_COLLECT_RANGE: 5419 return f2fs_compat_ioc_gc_range(file, arg); 5420 case F2FS_IOC32_MOVE_RANGE: 5421 return f2fs_compat_ioc_move_range(file, arg); 5422 case F2FS_IOC_START_ATOMIC_WRITE: 5423 case F2FS_IOC_START_ATOMIC_REPLACE: 5424 case F2FS_IOC_COMMIT_ATOMIC_WRITE: 5425 case F2FS_IOC_START_VOLATILE_WRITE: 5426 case F2FS_IOC_RELEASE_VOLATILE_WRITE: 5427 case F2FS_IOC_ABORT_ATOMIC_WRITE: 5428 case F2FS_IOC_SHUTDOWN: 5429 case FITRIM: 5430 case FS_IOC_SET_ENCRYPTION_POLICY: 5431 case FS_IOC_GET_ENCRYPTION_PWSALT: 5432 case FS_IOC_GET_ENCRYPTION_POLICY: 5433 case FS_IOC_GET_ENCRYPTION_POLICY_EX: 5434 case FS_IOC_ADD_ENCRYPTION_KEY: 5435 case FS_IOC_REMOVE_ENCRYPTION_KEY: 5436 case FS_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS: 5437 case FS_IOC_GET_ENCRYPTION_KEY_STATUS: 5438 case FS_IOC_GET_ENCRYPTION_NONCE: 5439 case F2FS_IOC_GARBAGE_COLLECT: 5440 case F2FS_IOC_WRITE_CHECKPOINT: 5441 case F2FS_IOC_DEFRAGMENT: 5442 case F2FS_IOC_FLUSH_DEVICE: 5443 case F2FS_IOC_GET_FEATURES: 5444 case F2FS_IOC_GET_PIN_FILE: 5445 case F2FS_IOC_SET_PIN_FILE: 5446 case F2FS_IOC_PRECACHE_EXTENTS: 5447 case F2FS_IOC_RESIZE_FS: 5448 case FS_IOC_ENABLE_VERITY: 5449 case FS_IOC_MEASURE_VERITY: 5450 case FS_IOC_READ_VERITY_METADATA: 5451 case FS_IOC_GETFSLABEL: 5452 case FS_IOC_SETFSLABEL: 5453 case F2FS_IOC_GET_COMPRESS_BLOCKS: 5454 case F2FS_IOC_RELEASE_COMPRESS_BLOCKS: 5455 case F2FS_IOC_RESERVE_COMPRESS_BLOCKS: 5456 case F2FS_IOC_SEC_TRIM_FILE: 5457 case F2FS_IOC_GET_COMPRESS_OPTION: 5458 case F2FS_IOC_SET_COMPRESS_OPTION: 5459 case F2FS_IOC_DECOMPRESS_FILE: 5460 case F2FS_IOC_COMPRESS_FILE: 5461 case F2FS_IOC_GET_DEV_ALIAS_FILE: 5462 case F2FS_IOC_IO_PRIO: 5463 break; 5464 default: 5465 return -ENOIOCTLCMD; 5466 } 5467 return __f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 5468 } 5469 #endif 5470 5471 const struct file_operations f2fs_file_operations = { 5472 .llseek = f2fs_llseek, 5473 .read_iter = f2fs_file_read_iter, 5474 .write_iter = f2fs_file_write_iter, 5475 .iopoll = iocb_bio_iopoll, 5476 .open = f2fs_file_open, 5477 .release = f2fs_release_file, 5478 .mmap_prepare = f2fs_file_mmap_prepare, 5479 .flush = f2fs_file_flush, 5480 .fsync = f2fs_sync_file, 5481 .fallocate = f2fs_fallocate, 5482 .unlocked_ioctl = f2fs_ioctl, 5483 #ifdef CONFIG_COMPAT 5484 .compat_ioctl = f2fs_compat_ioctl, 5485 #endif 5486 .splice_read = f2fs_file_splice_read, 5487 .splice_write = iter_file_splice_write, 5488 .fadvise = f2fs_file_fadvise, 5489 .fop_flags = FOP_BUFFER_RASYNC, 5490 .setlease = generic_setlease, 5491 }; 5492