fs/f2fs/file.c

   1 /*
   2  * fs/f2fs/file.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/stat.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/writeback.h>
  16 #include <linux/blkdev.h>
  17 #include <linux/falloc.h>
  18 #include <linux/types.h>
  19 #include <linux/compat.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/mount.h>
  22 #include <linux/pagevec.h>
  23
  24 #include "f2fs.h"
  25 #include "node.h"
  26 #include "segment.h"
  27 #include "xattr.h"
  28 #include "acl.h"
  29 #include "trace.h"
  30 #include <trace/events/f2fs.h>
  31
  32 static int f2fs_vm_page_mkwrite(struct vm_area_struct *vma,
  33                                                 struct vm_fault *vmf)
  34 {
  35         struct page *page = vmf->page;
  36         struct inode *inode = file_inode(vma->vm_file);
  37         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  38         struct dnode_of_data dn;
  39         int err;
  40
  41         f2fs_balance_fs(sbi);
  42
  43         sb_start_pagefault(inode->i_sb);
  44
  45         f2fs_bug_on(sbi, f2fs_has_inline_data(inode));
  46
  47         /* block allocation */
  48         f2fs_lock_op(sbi);
  49         set_new_dnode(&dn, inode, NULL, NULL, 0);
  50         err = f2fs_reserve_block(&dn, page->index);
  51         if (err) {
  52                 f2fs_unlock_op(sbi);
  53                 goto out;
  54         }
  55         f2fs_put_dnode(&dn);
  56         f2fs_unlock_op(sbi);
  57
  58         file_update_time(vma->vm_file);
  59         lock_page(page);
  60         if (unlikely(page->mapping != inode->i_mapping ||
  61                         page_offset(page) > i_size_read(inode) ||
  62                         !PageUptodate(page))) {
  63                 unlock_page(page);
  64                 err = -EFAULT;
  65                 goto out;
  66         }
  67
  68         /*
  69          * check to see if the page is mapped already (no holes)
  70          */
  71         if (PageMappedToDisk(page))
  72                 goto mapped;
  73
  74         /* page is wholly or partially inside EOF */
  75         if (((page->index + 1) << PAGE_CACHE_SHIFT) > i_size_read(inode)) {
  76                 unsigned offset;
  77                 offset = i_size_read(inode) & ~PAGE_CACHE_MASK;
  78                 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
  79         }
  80         set_page_dirty(page);
  81         SetPageUptodate(page);
  82
  83         trace_f2fs_vm_page_mkwrite(page, DATA);
  84 mapped:
  85         /* fill the page */
  86         f2fs_wait_on_page_writeback(page, DATA);
  87 out:
  88         sb_end_pagefault(inode->i_sb);
  89         return block_page_mkwrite_return(err);
  90 }
  91
  92 static const struct vm_operations_struct f2fs_file_vm_ops = {
  93         .fault          = filemap_fault,
  94         .map_pages      = filemap_map_pages,
  95         .page_mkwrite   = f2fs_vm_page_mkwrite,
  96 };
  97
  98 static int get_parent_ino(struct inode *inode, nid_t *pino)
  99 {
 100         struct dentry *dentry;
 101
 102         inode = igrab(inode);
 103         dentry = d_find_any_alias(inode);
 104         iput(inode);
 105         if (!dentry)
 106                 return 0;
 107
 108         if (update_dent_inode(inode, &dentry->d_name)) {
 109                 dput(dentry);
 110                 return 0;
 111         }
 112
 113         *pino = parent_ino(dentry);
 114         dput(dentry);
 115         return 1;
 116 }
 117
 118 static inline bool need_do_checkpoint(struct inode *inode)
 119 {
 120         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 121         bool need_cp = false;
 122
 123         if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1)
 124                 need_cp = true;
 125         else if (file_wrong_pino(inode))
 126                 need_cp = true;
 127         else if (!space_for_roll_forward(sbi))
 128                 need_cp = true;
 129         else if (!is_checkpointed_node(sbi, F2FS_I(inode)->i_pino))
 130                 need_cp = true;
 131         else if (F2FS_I(inode)->xattr_ver == cur_cp_version(F2FS_CKPT(sbi)))
 132                 need_cp = true;
 133         else if (test_opt(sbi, FASTBOOT))
 134                 need_cp = true;
 135         else if (sbi->active_logs == 2)
 136                 need_cp = true;
 137
 138         return need_cp;
 139 }
 140
 141 static bool need_inode_page_update(struct f2fs_sb_info *sbi, nid_t ino)
 142 {
 143         struct page *i = find_get_page(NODE_MAPPING(sbi), ino);
 144         bool ret = false;
 145         /* But we need to avoid that there are some inode updates */
 146         if ((i && PageDirty(i)) || need_inode_block_update(sbi, ino))
 147                 ret = true;
 148         f2fs_put_page(i, 0);
 149         return ret;
 150 }
 151
 152 static void try_to_fix_pino(struct inode *inode)
 153 {
 154         struct f2fs_inode_info *fi = F2FS_I(inode);
 155         nid_t pino;
 156
 157         down_write(&fi->i_sem);
 158         fi->xattr_ver = 0;
 159         if (file_wrong_pino(inode) && inode->i_nlink == 1 &&
 160                         get_parent_ino(inode, &pino)) {
 161                 fi->i_pino = pino;
 162                 file_got_pino(inode);
 163                 up_write(&fi->i_sem);
 164
 165                 mark_inode_dirty_sync(inode);
 166                 f2fs_write_inode(inode, NULL);
 167         } else {
 168                 up_write(&fi->i_sem);
 169         }
 170 }
 171
 172 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 173 {
 174         struct inode *inode = file->f_mapping->host;
 175         struct f2fs_inode_info *fi = F2FS_I(inode);
 176         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 177         nid_t ino = inode->i_ino;
 178         int ret = 0;
 179         bool need_cp = false;
 180         struct writeback_control wbc = {
 181                 .sync_mode = WB_SYNC_ALL,
 182                 .nr_to_write = LONG_MAX,
 183                 .for_reclaim = 0,
 184         };
 185
 186         if (unlikely(f2fs_readonly(inode->i_sb)))
 187                 return 0;
 188
 189         trace_f2fs_sync_file_enter(inode);
 190
 191         /* if fdatasync is triggered, let's do in-place-update */
 192         if (get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks)
 193                 set_inode_flag(fi, FI_NEED_IPU);
 194         ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
 195         clear_inode_flag(fi, FI_NEED_IPU);
 196
 197         if (ret) {
 198                 trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
 199                 return ret;
 200         }
 201
 202         /* if the inode is dirty, let's recover all the time */
 203         if (!datasync && is_inode_flag_set(fi, FI_DIRTY_INODE)) {
 204                 update_inode_page(inode);
 205                 goto go_write;
 206         }
 207
 208         /*
 209          * if there is no written data, don't waste time to write recovery info.
 210          */
 211         if (!is_inode_flag_set(fi, FI_APPEND_WRITE) &&
 212                         !exist_written_data(sbi, ino, APPEND_INO)) {
 213
 214                 /* it may call write_inode just prior to fsync */
 215                 if (need_inode_page_update(sbi, ino))
 216                         goto go_write;
 217
 218                 if (is_inode_flag_set(fi, FI_UPDATE_WRITE) ||
 219                                 exist_written_data(sbi, ino, UPDATE_INO))
 220                         goto flush_out;
 221                 goto out;
 222         }
 223 go_write:
 224         /* guarantee free sections for fsync */
 225         f2fs_balance_fs(sbi);
 226
 227         /*
 228          * Both of fdatasync() and fsync() are able to be recovered from
 229          * sudden-power-off.
 230          */
 231         down_read(&fi->i_sem);
 232         need_cp = need_do_checkpoint(inode);
 233         up_read(&fi->i_sem);
 234
 235         if (need_cp) {
 236                 /* all the dirty node pages should be flushed for POR */
 237                 ret = f2fs_sync_fs(inode->i_sb, 1);
 238
 239                 /*
 240                  * We've secured consistency through sync_fs. Following pino
 241                  * will be used only for fsynced inodes after checkpoint.
 242                  */
 243                 try_to_fix_pino(inode);
 244                 clear_inode_flag(fi, FI_APPEND_WRITE);
 245                 clear_inode_flag(fi, FI_UPDATE_WRITE);
 246                 goto out;
 247         }
 248 sync_nodes:
 249         sync_node_pages(sbi, ino, &wbc);
 250
 251         /* if cp_error was enabled, we should avoid infinite loop */
 252         if (unlikely(f2fs_cp_error(sbi)))
 253                 goto out;
 254
 255         if (need_inode_block_update(sbi, ino)) {
 256                 mark_inode_dirty_sync(inode);
 257                 f2fs_write_inode(inode, NULL);
 258                 goto sync_nodes;
 259         }
 260
 261         ret = wait_on_node_pages_writeback(sbi, ino);
 262         if (ret)
 263                 goto out;
 264
 265         /* once recovery info is written, don't need to tack this */
 266         remove_dirty_inode(sbi, ino, APPEND_INO);
 267         clear_inode_flag(fi, FI_APPEND_WRITE);
 268 flush_out:
 269         remove_dirty_inode(sbi, ino, UPDATE_INO);
 270         clear_inode_flag(fi, FI_UPDATE_WRITE);
 271         ret = f2fs_issue_flush(sbi);
 272 out:
 273         trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
 274         f2fs_trace_ios(NULL, NULL, 1);
 275         return ret;
 276 }
 277
 278 static pgoff_t __get_first_dirty_index(struct address_space *mapping,
 279                                                 pgoff_t pgofs, int whence)
 280 {
 281         struct pagevec pvec;
 282         int nr_pages;
 283
 284         if (whence != SEEK_DATA)
 285                 return 0;
 286
 287         /* find first dirty page index */
 288         pagevec_init(&pvec, 0);
 289         nr_pages = pagevec_lookup_tag(&pvec, mapping, &pgofs,
 290                                         PAGECACHE_TAG_DIRTY, 1);
 291         pgofs = nr_pages ? pvec.pages[0]->index : LONG_MAX;
 292         pagevec_release(&pvec);
 293         return pgofs;
 294 }
 295
 296 static bool __found_offset(block_t blkaddr, pgoff_t dirty, pgoff_t pgofs,
 297                                                         int whence)
 298 {
 299         switch (whence) {
 300         case SEEK_DATA:
 301                 if ((blkaddr == NEW_ADDR && dirty == pgofs) ||
 302                         (blkaddr != NEW_ADDR && blkaddr != NULL_ADDR))
 303                         return true;
 304                 break;
 305         case SEEK_HOLE:
 306                 if (blkaddr == NULL_ADDR)
 307                         return true;
 308                 break;
 309         }
 310         return false;
 311 }
 312
 313 static loff_t f2fs_seek_block(struct file *file, loff_t offset, int whence)
 314 {
 315         struct inode *inode = file->f_mapping->host;
 316         loff_t maxbytes = inode->i_sb->s_maxbytes;
 317         struct dnode_of_data dn;
 318         pgoff_t pgofs, end_offset, dirty;
 319         loff_t data_ofs = offset;
 320         loff_t isize;
 321         int err = 0;
 322
 323         mutex_lock(&inode->i_mutex);
 324
 325         isize = i_size_read(inode);
 326         if (offset >= isize)
 327                 goto fail;
 328
 329         /* handle inline data case */
 330         if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
 331                 if (whence == SEEK_HOLE)
 332                         data_ofs = isize;
 333                 goto found;
 334         }
 335
 336         pgofs = (pgoff_t)(offset >> PAGE_CACHE_SHIFT);
 337
 338         dirty = __get_first_dirty_index(inode->i_mapping, pgofs, whence);
 339
 340         for (; data_ofs < isize; data_ofs = pgofs << PAGE_CACHE_SHIFT) {
 341                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 342                 err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
 343                 if (err && err != -ENOENT) {
 344                         goto fail;
 345                 } else if (err == -ENOENT) {
 346                         /* direct node does not exists */
 347                         if (whence == SEEK_DATA) {
 348                                 pgofs = PGOFS_OF_NEXT_DNODE(pgofs,
 349                                                         F2FS_I(inode));
 350                                 continue;
 351                         } else {
 352                                 goto found;
 353                         }
 354                 }
 355
 356                 end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 357
 358                 /* find data/hole in dnode block */
 359                 for (; dn.ofs_in_node < end_offset;
 360                                 dn.ofs_in_node++, pgofs++,
 361                                 data_ofs = (loff_t)pgofs << PAGE_CACHE_SHIFT) {
 362                         block_t blkaddr;
 363                         blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
 364
 365                         if (__found_offset(blkaddr, dirty, pgofs, whence)) {
 366                                 f2fs_put_dnode(&dn);
 367                                 goto found;
 368                         }
 369                 }
 370                 f2fs_put_dnode(&dn);
 371         }
 372
 373         if (whence == SEEK_DATA)
 374                 goto fail;
 375 found:
 376         if (whence == SEEK_HOLE && data_ofs > isize)
 377                 data_ofs = isize;
 378         mutex_unlock(&inode->i_mutex);
 379         return vfs_setpos(file, data_ofs, maxbytes);
 380 fail:
 381         mutex_unlock(&inode->i_mutex);
 382         return -ENXIO;
 383 }
 384
 385 static loff_t f2fs_llseek(struct file *file, loff_t offset, int whence)
 386 {
 387         struct inode *inode = file->f_mapping->host;
 388         loff_t maxbytes = inode->i_sb->s_maxbytes;
 389
 390         switch (whence) {
 391         case SEEK_SET:
 392         case SEEK_CUR:
 393         case SEEK_END:
 394                 return generic_file_llseek_size(file, offset, whence,
 395                                                 maxbytes, i_size_read(inode));
 396         case SEEK_DATA:
 397         case SEEK_HOLE:
 398                 if (offset < 0)
 399                         return -ENXIO;
 400                 return f2fs_seek_block(file, offset, whence);
 401         }
 402
 403         return -EINVAL;
 404 }
 405
 406 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma)
 407 {
 408         struct inode *inode = file_inode(file);
 409
 410         /* we don't need to use inline_data strictly */
 411         if (f2fs_has_inline_data(inode)) {
 412                 int err = f2fs_convert_inline_inode(inode);
 413                 if (err)
 414                         return err;
 415         }
 416
 417         file_accessed(file);
 418         vma->vm_ops = &f2fs_file_vm_ops;
 419         return 0;
 420 }
 421
 422 int truncate_data_blocks_range(struct dnode_of_data *dn, int count)
 423 {
 424         int nr_free = 0, ofs = dn->ofs_in_node;
 425         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 426         struct f2fs_node *raw_node;
 427         __le32 *addr;
 428
 429         raw_node = F2FS_NODE(dn->node_page);
 430         addr = blkaddr_in_node(raw_node) + ofs;
 431
 432         for (; count > 0; count--, addr++, dn->ofs_in_node++) {
 433                 block_t blkaddr = le32_to_cpu(*addr);
 434                 if (blkaddr == NULL_ADDR)
 435                         continue;
 436
 437                 dn->data_blkaddr = NULL_ADDR;
 438                 f2fs_update_extent_cache(dn);
 439                 invalidate_blocks(sbi, blkaddr);
 440                 nr_free++;
 441         }
 442         if (nr_free) {
 443                 dec_valid_block_count(sbi, dn->inode, nr_free);
 444                 set_page_dirty(dn->node_page);
 445                 sync_inode_page(dn);
 446         }
 447         dn->ofs_in_node = ofs;
 448
 449         trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid,
 450                                          dn->ofs_in_node, nr_free);
 451         return nr_free;
 452 }
 453
 454 void truncate_data_blocks(struct dnode_of_data *dn)
 455 {
 456         truncate_data_blocks_range(dn, ADDRS_PER_BLOCK);
 457 }
 458
 459 static int truncate_partial_data_page(struct inode *inode, u64 from)
 460 {
 461         unsigned offset = from & (PAGE_CACHE_SIZE - 1);
 462         struct page *page;
 463
 464         if (!offset)
 465                 return 0;
 466
 467         page = find_data_page(inode, from >> PAGE_CACHE_SHIFT, false);
 468         if (IS_ERR(page))
 469                 return 0;
 470
 471         lock_page(page);
 472         if (unlikely(!PageUptodate(page) ||
 473                         page->mapping != inode->i_mapping))
 474                 goto out;
 475
 476         f2fs_wait_on_page_writeback(page, DATA);
 477         zero_user(page, offset, PAGE_CACHE_SIZE - offset);
 478         set_page_dirty(page);
 479 out:
 480         f2fs_put_page(page, 1);
 481         return 0;
 482 }
 483
 484 int truncate_blocks(struct inode *inode, u64 from, bool lock)
 485 {
 486         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 487         unsigned int blocksize = inode->i_sb->s_blocksize;
 488         struct dnode_of_data dn;
 489         pgoff_t free_from;
 490         int count = 0, err = 0;
 491         struct page *ipage;
 492
 493         trace_f2fs_truncate_blocks_enter(inode, from);
 494
 495         free_from = (pgoff_t)F2FS_BYTES_TO_BLK(from + blocksize - 1);
 496
 497         if (lock)
 498                 f2fs_lock_op(sbi);
 499
 500         ipage = get_node_page(sbi, inode->i_ino);
 501         if (IS_ERR(ipage)) {
 502                 err = PTR_ERR(ipage);
 503                 goto out;
 504         }
 505
 506         if (f2fs_has_inline_data(inode)) {
 507                 f2fs_put_page(ipage, 1);
 508                 goto out;
 509         }
 510
 511         set_new_dnode(&dn, inode, ipage, NULL, 0);
 512         err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE);
 513         if (err) {
 514                 if (err == -ENOENT)
 515                         goto free_next;
 516                 goto out;
 517         }
 518
 519         count = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
 520
 521         count -= dn.ofs_in_node;
 522         f2fs_bug_on(sbi, count < 0);
 523
 524         if (dn.ofs_in_node || IS_INODE(dn.node_page)) {
 525                 truncate_data_blocks_range(&dn, count);
 526                 free_from += count;
 527         }
 528
 529         f2fs_put_dnode(&dn);
 530 free_next:
 531         err = truncate_inode_blocks(inode, free_from);
 532 out:
 533         if (lock)
 534                 f2fs_unlock_op(sbi);
 535
 536         /* lastly zero out the first data page */
 537         if (!err)
 538                 err = truncate_partial_data_page(inode, from);
 539
 540         trace_f2fs_truncate_blocks_exit(inode, err);
 541         return err;
 542 }
 543
 544 void f2fs_truncate(struct inode *inode)
 545 {
 546         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 547                                 S_ISLNK(inode->i_mode)))
 548                 return;
 549
 550         trace_f2fs_truncate(inode);
 551
 552         /* we should check inline_data size */
 553         if (f2fs_has_inline_data(inode) && !f2fs_may_inline(inode)) {
 554                 if (f2fs_convert_inline_inode(inode))
 555                         return;
 556         }
 557
 558         if (!truncate_blocks(inode, i_size_read(inode), true)) {
 559                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 560                 mark_inode_dirty(inode);
 561         }
 562 }
 563
 564 int f2fs_getattr(struct vfsmount *mnt,
 565                          struct dentry *dentry, struct kstat *stat)
 566 {
 567         struct inode *inode = dentry->d_inode;
 568         generic_fillattr(inode, stat);
 569         stat->blocks <<= 3;
 570         return 0;
 571 }
 572
 573 #ifdef CONFIG_F2FS_FS_POSIX_ACL
 574 static void __setattr_copy(struct inode *inode, const struct iattr *attr)
 575 {
 576         struct f2fs_inode_info *fi = F2FS_I(inode);
 577         unsigned int ia_valid = attr->ia_valid;
 578
 579         if (ia_valid & ATTR_UID)
 580                 inode->i_uid = attr->ia_uid;
 581         if (ia_valid & ATTR_GID)
 582                 inode->i_gid = attr->ia_gid;
 583         if (ia_valid & ATTR_ATIME)
 584                 inode->i_atime = timespec_trunc(attr->ia_atime,
 585                                                 inode->i_sb->s_time_gran);
 586         if (ia_valid & ATTR_MTIME)
 587                 inode->i_mtime = timespec_trunc(attr->ia_mtime,
 588                                                 inode->i_sb->s_time_gran);
 589         if (ia_valid & ATTR_CTIME)
 590                 inode->i_ctime = timespec_trunc(attr->ia_ctime,
 591                                                 inode->i_sb->s_time_gran);
 592         if (ia_valid & ATTR_MODE) {
 593                 umode_t mode = attr->ia_mode;
 594
 595                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 596                         mode &= ~S_ISGID;
 597                 set_acl_inode(fi, mode);
 598         }
 599 }
 600 #else
 601 #define __setattr_copy setattr_copy
 602 #endif
 603
 604 int f2fs_setattr(struct dentry *dentry, struct iattr *attr)
 605 {
 606         struct inode *inode = dentry->d_inode;
 607         struct f2fs_inode_info *fi = F2FS_I(inode);
 608         int err;
 609
 610         err = inode_change_ok(inode, attr);
 611         if (err)
 612                 return err;
 613
 614         if (attr->ia_valid & ATTR_SIZE) {
 615                 if (attr->ia_size != i_size_read(inode)) {
 616                         truncate_setsize(inode, attr->ia_size);
 617                         f2fs_truncate(inode);
 618                         f2fs_balance_fs(F2FS_I_SB(inode));
 619                 } else {
 620                         /*
 621                          * giving a chance to truncate blocks past EOF which
 622                          * are fallocated with FALLOC_FL_KEEP_SIZE.
 623                          */
 624                         f2fs_truncate(inode);
 625                 }
 626         }
 627
 628         __setattr_copy(inode, attr);
 629
 630         if (attr->ia_valid & ATTR_MODE) {
 631                 err = posix_acl_chmod(inode, get_inode_mode(inode));
 632                 if (err || is_inode_flag_set(fi, FI_ACL_MODE)) {
 633                         inode->i_mode = fi->i_acl_mode;
 634                         clear_inode_flag(fi, FI_ACL_MODE);
 635                 }
 636         }
 637
 638         mark_inode_dirty(inode);
 639         return err;
 640 }
 641
 642 const struct inode_operations f2fs_file_inode_operations = {
 643         .getattr        = f2fs_getattr,
 644         .setattr        = f2fs_setattr,
 645         .get_acl        = f2fs_get_acl,
 646         .set_acl        = f2fs_set_acl,
 647 #ifdef CONFIG_F2FS_FS_XATTR
 648         .setxattr       = generic_setxattr,
 649         .getxattr       = generic_getxattr,
 650         .listxattr      = f2fs_listxattr,
 651         .removexattr    = generic_removexattr,
 652 #endif
 653         .fiemap         = f2fs_fiemap,
 654 };
 655
 656 static void fill_zero(struct inode *inode, pgoff_t index,
 657                                         loff_t start, loff_t len)
 658 {
 659         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 660         struct page *page;
 661
 662         if (!len)
 663                 return;
 664
 665         f2fs_balance_fs(sbi);
 666
 667         f2fs_lock_op(sbi);
 668         page = get_new_data_page(inode, NULL, index, false);
 669         f2fs_unlock_op(sbi);
 670
 671         if (!IS_ERR(page)) {
 672                 f2fs_wait_on_page_writeback(page, DATA);
 673                 zero_user(page, start, len);
 674                 set_page_dirty(page);
 675                 f2fs_put_page(page, 1);
 676         }
 677 }
 678
 679 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end)
 680 {
 681         pgoff_t index;
 682         int err;
 683
 684         for (index = pg_start; index < pg_end; index++) {
 685                 struct dnode_of_data dn;
 686
 687                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 688                 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
 689                 if (err) {
 690                         if (err == -ENOENT)
 691                                 continue;
 692                         return err;
 693                 }
 694
 695                 if (dn.data_blkaddr != NULL_ADDR)
 696                         truncate_data_blocks_range(&dn, 1);
 697                 f2fs_put_dnode(&dn);
 698         }
 699         return 0;
 700 }
 701
 702 static int punch_hole(struct inode *inode, loff_t offset, loff_t len)
 703 {
 704         pgoff_t pg_start, pg_end;
 705         loff_t off_start, off_end;
 706         int ret = 0;
 707
 708         if (!S_ISREG(inode->i_mode))
 709                 return -EOPNOTSUPP;
 710
 711         /* skip punching hole beyond i_size */
 712         if (offset >= inode->i_size)
 713                 return ret;
 714
 715         if (f2fs_has_inline_data(inode)) {
 716                 ret = f2fs_convert_inline_inode(inode);
 717                 if (ret)
 718                         return ret;
 719         }
 720
 721         pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
 722         pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
 723
 724         off_start = offset & (PAGE_CACHE_SIZE - 1);
 725         off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
 726
 727         if (pg_start == pg_end) {
 728                 fill_zero(inode, pg_start, off_start,
 729                                                 off_end - off_start);
 730         } else {
 731                 if (off_start)
 732                         fill_zero(inode, pg_start++, off_start,
 733                                         PAGE_CACHE_SIZE - off_start);
 734                 if (off_end)
 735                         fill_zero(inode, pg_end, 0, off_end);
 736
 737                 if (pg_start < pg_end) {
 738                         struct address_space *mapping = inode->i_mapping;
 739                         loff_t blk_start, blk_end;
 740                         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 741
 742                         f2fs_balance_fs(sbi);
 743
 744                         blk_start = pg_start << PAGE_CACHE_SHIFT;
 745                         blk_end = pg_end << PAGE_CACHE_SHIFT;
 746                         truncate_inode_pages_range(mapping, blk_start,
 747                                         blk_end - 1);
 748
 749                         f2fs_lock_op(sbi);
 750                         ret = truncate_hole(inode, pg_start, pg_end);
 751                         f2fs_unlock_op(sbi);
 752                 }
 753         }
 754
 755         return ret;
 756 }
 757
 758 static int expand_inode_data(struct inode *inode, loff_t offset,
 759                                         loff_t len, int mode)
 760 {
 761         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 762         pgoff_t index, pg_start, pg_end;
 763         loff_t new_size = i_size_read(inode);
 764         loff_t off_start, off_end;
 765         int ret = 0;
 766
 767         f2fs_balance_fs(sbi);
 768
 769         ret = inode_newsize_ok(inode, (len + offset));
 770         if (ret)
 771                 return ret;
 772
 773         if (f2fs_has_inline_data(inode)) {
 774                 ret = f2fs_convert_inline_inode(inode);
 775                 if (ret)
 776                         return ret;
 777         }
 778
 779         pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
 780         pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
 781
 782         off_start = offset & (PAGE_CACHE_SIZE - 1);
 783         off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
 784
 785         f2fs_lock_op(sbi);
 786
 787         for (index = pg_start; index <= pg_end; index++) {
 788                 struct dnode_of_data dn;
 789
 790                 if (index == pg_end && !off_end)
 791                         goto noalloc;
 792
 793                 set_new_dnode(&dn, inode, NULL, NULL, 0);
 794                 ret = f2fs_reserve_block(&dn, index);
 795                 if (ret)
 796                         break;
 797 noalloc:
 798                 if (pg_start == pg_end)
 799                         new_size = offset + len;
 800                 else if (index == pg_start && off_start)
 801                         new_size = (index + 1) << PAGE_CACHE_SHIFT;
 802                 else if (index == pg_end)
 803                         new_size = (index << PAGE_CACHE_SHIFT) + off_end;
 804                 else
 805                         new_size += PAGE_CACHE_SIZE;
 806         }
 807
 808         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
 809                 i_size_read(inode) < new_size) {
 810                 i_size_write(inode, new_size);
 811                 mark_inode_dirty(inode);
 812                 update_inode_page(inode);
 813         }
 814         f2fs_unlock_op(sbi);
 815
 816         return ret;
 817 }
 818
 819 static long f2fs_fallocate(struct file *file, int mode,
 820                                 loff_t offset, loff_t len)
 821 {
 822         struct inode *inode = file_inode(file);
 823         long ret;
 824
 825         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 826                 return -EOPNOTSUPP;
 827
 828         mutex_lock(&inode->i_mutex);
 829
 830         if (mode & FALLOC_FL_PUNCH_HOLE)
 831                 ret = punch_hole(inode, offset, len);
 832         else
 833                 ret = expand_inode_data(inode, offset, len, mode);
 834
 835         if (!ret) {
 836                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 837                 mark_inode_dirty(inode);
 838         }
 839
 840         mutex_unlock(&inode->i_mutex);
 841
 842         trace_f2fs_fallocate(inode, mode, offset, len, ret);
 843         return ret;
 844 }
 845
 846 static int f2fs_release_file(struct inode *inode, struct file *filp)
 847 {
 848         /* some remained atomic pages should discarded */
 849         if (f2fs_is_atomic_file(inode))
 850                 commit_inmem_pages(inode, true);
 851         if (f2fs_is_volatile_file(inode)) {
 852                 set_inode_flag(F2FS_I(inode), FI_DROP_CACHE);
 853                 filemap_fdatawrite(inode->i_mapping);
 854                 clear_inode_flag(F2FS_I(inode), FI_DROP_CACHE);
 855         }
 856         return 0;
 857 }
 858
 859 #define F2FS_REG_FLMASK         (~(FS_DIRSYNC_FL | FS_TOPDIR_FL))
 860 #define F2FS_OTHER_FLMASK       (FS_NODUMP_FL | FS_NOATIME_FL)
 861
 862 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags)
 863 {
 864         if (S_ISDIR(mode))
 865                 return flags;
 866         else if (S_ISREG(mode))
 867                 return flags & F2FS_REG_FLMASK;
 868         else
 869                 return flags & F2FS_OTHER_FLMASK;
 870 }
 871
 872 static int f2fs_ioc_getflags(struct file *filp, unsigned long arg)
 873 {
 874         struct inode *inode = file_inode(filp);
 875         struct f2fs_inode_info *fi = F2FS_I(inode);
 876         unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
 877         return put_user(flags, (int __user *)arg);
 878 }
 879
 880 static int f2fs_ioc_setflags(struct file *filp, unsigned long arg)
 881 {
 882         struct inode *inode = file_inode(filp);
 883         struct f2fs_inode_info *fi = F2FS_I(inode);
 884         unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
 885         unsigned int oldflags;
 886         int ret;
 887
 888         ret = mnt_want_write_file(filp);
 889         if (ret)
 890                 return ret;
 891
 892         if (!inode_owner_or_capable(inode)) {
 893                 ret = -EACCES;
 894                 goto out;
 895         }
 896
 897         if (get_user(flags, (int __user *)arg)) {
 898                 ret = -EFAULT;
 899                 goto out;
 900         }
 901
 902         flags = f2fs_mask_flags(inode->i_mode, flags);
 903
 904         mutex_lock(&inode->i_mutex);
 905
 906         oldflags = fi->i_flags;
 907
 908         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 909                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 910                         mutex_unlock(&inode->i_mutex);
 911                         ret = -EPERM;
 912                         goto out;
 913                 }
 914         }
 915
 916         flags = flags & FS_FL_USER_MODIFIABLE;
 917         flags |= oldflags & ~FS_FL_USER_MODIFIABLE;
 918         fi->i_flags = flags;
 919         mutex_unlock(&inode->i_mutex);
 920
 921         f2fs_set_inode_flags(inode);
 922         inode->i_ctime = CURRENT_TIME;
 923         mark_inode_dirty(inode);
 924 out:
 925         mnt_drop_write_file(filp);
 926         return ret;
 927 }
 928
 929 static int f2fs_ioc_getversion(struct file *filp, unsigned long arg)
 930 {
 931         struct inode *inode = file_inode(filp);
 932
 933         return put_user(inode->i_generation, (int __user *)arg);
 934 }
 935
 936 static int f2fs_ioc_start_atomic_write(struct file *filp)
 937 {
 938         struct inode *inode = file_inode(filp);
 939
 940         if (!inode_owner_or_capable(inode))
 941                 return -EACCES;
 942
 943         f2fs_balance_fs(F2FS_I_SB(inode));
 944
 945         if (f2fs_is_atomic_file(inode))
 946                 return 0;
 947
 948         set_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
 949
 950         return f2fs_convert_inline_inode(inode);
 951 }
 952
 953 static int f2fs_ioc_commit_atomic_write(struct file *filp)
 954 {
 955         struct inode *inode = file_inode(filp);
 956         int ret;
 957
 958         if (!inode_owner_or_capable(inode))
 959                 return -EACCES;
 960
 961         if (f2fs_is_volatile_file(inode))
 962                 return 0;
 963
 964         ret = mnt_want_write_file(filp);
 965         if (ret)
 966                 return ret;
 967
 968         if (f2fs_is_atomic_file(inode))
 969                 commit_inmem_pages(inode, false);
 970
 971         ret = f2fs_sync_file(filp, 0, LONG_MAX, 0);
 972         mnt_drop_write_file(filp);
 973         clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
 974         return ret;
 975 }
 976
 977 static int f2fs_ioc_start_volatile_write(struct file *filp)
 978 {
 979         struct inode *inode = file_inode(filp);
 980
 981         if (!inode_owner_or_capable(inode))
 982                 return -EACCES;
 983
 984         if (f2fs_is_volatile_file(inode))
 985                 return 0;
 986
 987         set_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
 988
 989         return f2fs_convert_inline_inode(inode);
 990 }
 991
 992 static int f2fs_ioc_release_volatile_write(struct file *filp)
 993 {
 994         struct inode *inode = file_inode(filp);
 995
 996         if (!inode_owner_or_capable(inode))
 997                 return -EACCES;
 998
 999         if (!f2fs_is_volatile_file(inode))
1000                 return 0;
1001
1002         punch_hole(inode, 0, F2FS_BLKSIZE);
1003         return 0;
1004 }
1005
1006 static int f2fs_ioc_abort_volatile_write(struct file *filp)
1007 {
1008         struct inode *inode = file_inode(filp);
1009         int ret;
1010
1011         if (!inode_owner_or_capable(inode))
1012                 return -EACCES;
1013
1014         ret = mnt_want_write_file(filp);
1015         if (ret)
1016                 return ret;
1017
1018         f2fs_balance_fs(F2FS_I_SB(inode));
1019
1020         if (f2fs_is_atomic_file(inode)) {
1021                 commit_inmem_pages(inode, false);
1022                 clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
1023         }
1024
1025         if (f2fs_is_volatile_file(inode)) {
1026                 clear_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
1027                 filemap_fdatawrite(inode->i_mapping);
1028                 set_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
1029         }
1030         mnt_drop_write_file(filp);
1031         return ret;
1032 }
1033
1034 static int f2fs_ioc_shutdown(struct file *filp, unsigned long arg)
1035 {
1036         struct inode *inode = file_inode(filp);
1037         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1038         struct super_block *sb = sbi->sb;
1039         __u32 in;
1040
1041         if (!capable(CAP_SYS_ADMIN))
1042                 return -EPERM;
1043
1044         if (get_user(in, (__u32 __user *)arg))
1045                 return -EFAULT;
1046
1047         switch (in) {
1048         case F2FS_GOING_DOWN_FULLSYNC:
1049                 sb = freeze_bdev(sb->s_bdev);
1050                 if (sb && !IS_ERR(sb)) {
1051                         f2fs_stop_checkpoint(sbi);
1052                         thaw_bdev(sb->s_bdev, sb);
1053                 }
1054                 break;
1055         case F2FS_GOING_DOWN_METASYNC:
1056                 /* do checkpoint only */
1057                 f2fs_sync_fs(sb, 1);
1058                 f2fs_stop_checkpoint(sbi);
1059                 break;
1060         case F2FS_GOING_DOWN_NOSYNC:
1061                 f2fs_stop_checkpoint(sbi);
1062                 break;
1063         default:
1064                 return -EINVAL;
1065         }
1066         return 0;
1067 }
1068
1069 static int f2fs_ioc_fitrim(struct file *filp, unsigned long arg)
1070 {
1071         struct inode *inode = file_inode(filp);
1072         struct super_block *sb = inode->i_sb;
1073         struct request_queue *q = bdev_get_queue(sb->s_bdev);
1074         struct fstrim_range range;
1075         int ret;
1076
1077         if (!capable(CAP_SYS_ADMIN))
1078                 return -EPERM;
1079
1080         if (!blk_queue_discard(q))
1081                 return -EOPNOTSUPP;
1082
1083         if (copy_from_user(&range, (struct fstrim_range __user *)arg,
1084                                 sizeof(range)))
1085                 return -EFAULT;
1086
1087         range.minlen = max((unsigned int)range.minlen,
1088                                 q->limits.discard_granularity);
1089         ret = f2fs_trim_fs(F2FS_SB(sb), &range);
1090         if (ret < 0)
1091                 return ret;
1092
1093         if (copy_to_user((struct fstrim_range __user *)arg, &range,
1094                                 sizeof(range)))
1095                 return -EFAULT;
1096         return 0;
1097 }
1098
1099 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1100 {
1101         switch (cmd) {
1102         case F2FS_IOC_GETFLAGS:
1103                 return f2fs_ioc_getflags(filp, arg);
1104         case F2FS_IOC_SETFLAGS:
1105                 return f2fs_ioc_setflags(filp, arg);
1106         case F2FS_IOC_GETVERSION:
1107                 return f2fs_ioc_getversion(filp, arg);
1108         case F2FS_IOC_START_ATOMIC_WRITE:
1109                 return f2fs_ioc_start_atomic_write(filp);
1110         case F2FS_IOC_COMMIT_ATOMIC_WRITE:
1111                 return f2fs_ioc_commit_atomic_write(filp);
1112         case F2FS_IOC_START_VOLATILE_WRITE:
1113                 return f2fs_ioc_start_volatile_write(filp);
1114         case F2FS_IOC_RELEASE_VOLATILE_WRITE:
1115                 return f2fs_ioc_release_volatile_write(filp);
1116         case F2FS_IOC_ABORT_VOLATILE_WRITE:
1117                 return f2fs_ioc_abort_volatile_write(filp);
1118         case F2FS_IOC_SHUTDOWN:
1119                 return f2fs_ioc_shutdown(filp, arg);
1120         case FITRIM:
1121                 return f2fs_ioc_fitrim(filp, arg);
1122         default:
1123                 return -ENOTTY;
1124         }
1125 }
1126
1127 #ifdef CONFIG_COMPAT
1128 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1129 {
1130         switch (cmd) {
1131         case F2FS_IOC32_GETFLAGS:
1132                 cmd = F2FS_IOC_GETFLAGS;
1133                 break;
1134         case F2FS_IOC32_SETFLAGS:
1135                 cmd = F2FS_IOC_SETFLAGS;
1136                 break;
1137         default:
1138                 return -ENOIOCTLCMD;
1139         }
1140         return f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
1141 }
1142 #endif
1143
1144 const struct file_operations f2fs_file_operations = {
1145         .llseek         = f2fs_llseek,
1146         .read           = new_sync_read,
1147         .write          = new_sync_write,
1148         .read_iter      = generic_file_read_iter,
1149         .write_iter     = generic_file_write_iter,
1150         .open           = generic_file_open,
1151         .release        = f2fs_release_file,
1152         .mmap           = f2fs_file_mmap,
1153         .fsync          = f2fs_sync_file,
1154         .fallocate      = f2fs_fallocate,
1155         .unlocked_ioctl = f2fs_ioctl,
1156 #ifdef CONFIG_COMPAT
1157         .compat_ioctl   = f2fs_compat_ioctl,
1158 #endif
1159         .splice_read    = generic_file_splice_read,
1160         .splice_write   = iter_file_splice_write,
1161 };