fs/ext4/file.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/file.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/file.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  ext4 fs regular file handling primitives
  17  *
  18  *  64-bit file support on 64-bit platforms by Jakub Jelinek
  19  *      (jj@sunsite.ms.mff.cuni.cz)
  20  */
  21
  22 #include <linux/time.h>
  23 #include <linux/fs.h>
  24 #include <linux/iomap.h>
  25 #include <linux/mount.h>
  26 #include <linux/path.h>
  27 #include <linux/dax.h>
  28 #include <linux/quotaops.h>
  29 #include <linux/pagevec.h>
  30 #include <linux/uio.h>
  31 #include <linux/mman.h>
  32 #include <linux/backing-dev.h>
  33 #include "ext4.h"
  34 #include "ext4_jbd2.h"
  35 #include "xattr.h"
  36 #include "acl.h"
  37 #include "truncate.h"
  38
  39 static bool ext4_dio_supported(struct inode *inode)
  40 {
  41         if (IS_ENABLED(CONFIG_FS_ENCRYPTION) && IS_ENCRYPTED(inode))
  42                 return false;
  43         if (fsverity_active(inode))
  44                 return false;
  45         if (ext4_should_journal_data(inode))
  46                 return false;
  47         if (ext4_has_inline_data(inode))
  48                 return false;
  49         return true;
  50 }
  51
  52 static ssize_t ext4_dio_read_iter(struct kiocb *iocb, struct iov_iter *to)
  53 {
  54         ssize_t ret;
  55         struct inode *inode = file_inode(iocb->ki_filp);
  56
  57         if (iocb->ki_flags & IOCB_NOWAIT) {
  58                 if (!inode_trylock_shared(inode))
  59                         return -EAGAIN;
  60         } else {
  61                 inode_lock_shared(inode);
  62         }
  63
  64         if (!ext4_dio_supported(inode)) {
  65                 inode_unlock_shared(inode);
  66                 /*
  67                  * Fallback to buffered I/O if the operation being performed on
  68                  * the inode is not supported by direct I/O. The IOCB_DIRECT
  69                  * flag needs to be cleared here in order to ensure that the
  70                  * direct I/O path within generic_file_read_iter() is not
  71                  * taken.
  72                  */
  73                 iocb->ki_flags &= ~IOCB_DIRECT;
  74                 return generic_file_read_iter(iocb, to);
  75         }
  76
  77         ret = iomap_dio_rw(iocb, to, &ext4_iomap_ops, NULL,
  78                            is_sync_kiocb(iocb));
  79         inode_unlock_shared(inode);
  80
  81         file_accessed(iocb->ki_filp);
  82         return ret;
  83 }
  84
  85 #ifdef CONFIG_FS_DAX
  86 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
  87 {
  88         struct inode *inode = file_inode(iocb->ki_filp);
  89         ssize_t ret;
  90
  91         if (!inode_trylock_shared(inode)) {
  92                 if (iocb->ki_flags & IOCB_NOWAIT)
  93                         return -EAGAIN;
  94                 inode_lock_shared(inode);
  95         }
  96         /*
  97          * Recheck under inode lock - at this point we are sure it cannot
  98          * change anymore
  99          */
 100         if (!IS_DAX(inode)) {
 101                 inode_unlock_shared(inode);
 102                 /* Fallback to buffered IO in case we cannot support DAX */
 103                 return generic_file_read_iter(iocb, to);
 104         }
 105         ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
 106         inode_unlock_shared(inode);
 107
 108         file_accessed(iocb->ki_filp);
 109         return ret;
 110 }
 111 #endif
 112
 113 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 114 {
 115         struct inode *inode = file_inode(iocb->ki_filp);
 116
 117         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 118                 return -EIO;
 119
 120         if (!iov_iter_count(to))
 121                 return 0; /* skip atime */
 122
 123 #ifdef CONFIG_FS_DAX
 124         if (IS_DAX(inode))
 125                 return ext4_dax_read_iter(iocb, to);
 126 #endif
 127         if (iocb->ki_flags & IOCB_DIRECT)
 128                 return ext4_dio_read_iter(iocb, to);
 129
 130         return generic_file_read_iter(iocb, to);
 131 }
 132
 133 /*
 134  * Called when an inode is released. Note that this is different
 135  * from ext4_file_open: open gets called at every open, but release
 136  * gets called only when /all/ the files are closed.
 137  */
 138 static int ext4_release_file(struct inode *inode, struct file *filp)
 139 {
 140         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
 141                 ext4_alloc_da_blocks(inode);
 142                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
 143         }
 144         /* if we are the last writer on the inode, drop the block reservation */
 145         if ((filp->f_mode & FMODE_WRITE) &&
 146                         (atomic_read(&inode->i_writecount) == 1) &&
 147                         !EXT4_I(inode)->i_reserved_data_blocks)
 148         {
 149                 down_write(&EXT4_I(inode)->i_data_sem);
 150                 ext4_discard_preallocations(inode);
 151                 up_write(&EXT4_I(inode)->i_data_sem);
 152         }
 153         if (is_dx(inode) && filp->private_data)
 154                 ext4_htree_free_dir_info(filp->private_data);
 155
 156         return 0;
 157 }
 158
 159 /*
 160  * This tests whether the IO in question is block-aligned or not.
 161  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
 162  * are converted to written only after the IO is complete.  Until they are
 163  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
 164  * it needs to zero out portions of the start and/or end block.  If 2 AIO
 165  * threads are at work on the same unwritten block, they must be synchronized
 166  * or one thread will zero the other's data, causing corruption.
 167  */
 168 static int
 169 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
 170 {
 171         struct super_block *sb = inode->i_sb;
 172         int blockmask = sb->s_blocksize - 1;
 173
 174         if (pos >= ALIGN(i_size_read(inode), sb->s_blocksize))
 175                 return 0;
 176
 177         if ((pos | iov_iter_alignment(from)) & blockmask)
 178                 return 1;
 179
 180         return 0;
 181 }
 182
 183 /* Is IO overwriting allocated and initialized blocks? */
 184 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
 185 {
 186         struct ext4_map_blocks map;
 187         unsigned int blkbits = inode->i_blkbits;
 188         int err, blklen;
 189
 190         if (pos + len > i_size_read(inode))
 191                 return false;
 192
 193         map.m_lblk = pos >> blkbits;
 194         map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
 195         blklen = map.m_len;
 196
 197         err = ext4_map_blocks(NULL, inode, &map, 0);
 198         /*
 199          * 'err==len' means that all of the blocks have been preallocated,
 200          * regardless of whether they have been initialized or not. To exclude
 201          * unwritten extents, we need to check m_flags.
 202          */
 203         return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
 204 }
 205
 206 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
 207 {
 208         struct inode *inode = file_inode(iocb->ki_filp);
 209         ssize_t ret;
 210
 211         if (unlikely(IS_IMMUTABLE(inode)))
 212                 return -EPERM;
 213
 214         ret = generic_write_checks(iocb, from);
 215         if (ret <= 0)
 216                 return ret;
 217
 218         /*
 219          * If we have encountered a bitmap-format file, the size limit
 220          * is smaller than s_maxbytes, which is for extent-mapped files.
 221          */
 222         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
 223                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 224
 225                 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
 226                         return -EFBIG;
 227                 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
 228         }
 229
 230         ret = file_modified(iocb->ki_filp);
 231         if (ret)
 232                 return ret;
 233
 234         return iov_iter_count(from);
 235 }
 236
 237 static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
 238                                         struct iov_iter *from)
 239 {
 240         ssize_t ret;
 241         struct inode *inode = file_inode(iocb->ki_filp);
 242
 243         if (iocb->ki_flags & IOCB_NOWAIT)
 244                 return -EOPNOTSUPP;
 245
 246         inode_lock(inode);
 247         ret = ext4_write_checks(iocb, from);
 248         if (ret <= 0)
 249                 goto out;
 250
 251         current->backing_dev_info = inode_to_bdi(inode);
 252         ret = generic_perform_write(iocb->ki_filp, from, iocb->ki_pos);
 253         current->backing_dev_info = NULL;
 254
 255 out:
 256         inode_unlock(inode);
 257         if (likely(ret > 0)) {
 258                 iocb->ki_pos += ret;
 259                 ret = generic_write_sync(iocb, ret);
 260         }
 261
 262         return ret;
 263 }
 264
 265 static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
 266                                            ssize_t written, size_t count)
 267 {
 268         handle_t *handle;
 269         bool truncate = false;
 270         u8 blkbits = inode->i_blkbits;
 271         ext4_lblk_t written_blk, end_blk;
 272
 273         /*
 274          * Note that EXT4_I(inode)->i_disksize can get extended up to
 275          * inode->i_size while the I/O was running due to writeback of delalloc
 276          * blocks. But, the code in ext4_iomap_alloc() is careful to use
 277          * zeroed/unwritten extents if this is possible; thus we won't leave
 278          * uninitialized blocks in a file even if we didn't succeed in writing
 279          * as much as we intended.
 280          */
 281         WARN_ON_ONCE(i_size_read(inode) < EXT4_I(inode)->i_disksize);
 282         if (offset + count <= EXT4_I(inode)->i_disksize) {
 283                 /*
 284                  * We need to ensure that the inode is removed from the orphan
 285                  * list if it has been added prematurely, due to writeback of
 286                  * delalloc blocks.
 287                  */
 288                 if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) {
 289                         handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
 290
 291                         if (IS_ERR(handle)) {
 292                                 ext4_orphan_del(NULL, inode);
 293                                 return PTR_ERR(handle);
 294                         }
 295
 296                         ext4_orphan_del(handle, inode);
 297                         ext4_journal_stop(handle);
 298                 }
 299
 300                 return written;
 301         }
 302
 303         if (written < 0)
 304                 goto truncate;
 305
 306         handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
 307         if (IS_ERR(handle)) {
 308                 written = PTR_ERR(handle);
 309                 goto truncate;
 310         }
 311
 312         if (ext4_update_inode_size(inode, offset + written))
 313                 ext4_mark_inode_dirty(handle, inode);
 314
 315         /*
 316          * We may need to truncate allocated but not written blocks beyond EOF.
 317          */
 318         written_blk = ALIGN(offset + written, 1 << blkbits);
 319         end_blk = ALIGN(offset + count, 1 << blkbits);
 320         if (written_blk < end_blk && ext4_can_truncate(inode))
 321                 truncate = true;
 322
 323         /*
 324          * Remove the inode from the orphan list if it has been extended and
 325          * everything went OK.
 326          */
 327         if (!truncate && inode->i_nlink)
 328                 ext4_orphan_del(handle, inode);
 329         ext4_journal_stop(handle);
 330
 331         if (truncate) {
 332 truncate:
 333                 ext4_truncate_failed_write(inode);
 334                 /*
 335                  * If the truncate operation failed early, then the inode may
 336                  * still be on the orphan list. In that case, we need to try
 337                  * remove the inode from the in-memory linked list.
 338                  */
 339                 if (inode->i_nlink)
 340                         ext4_orphan_del(NULL, inode);
 341         }
 342
 343         return written;
 344 }
 345
 346 static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
 347                                  int error, unsigned int flags)
 348 {
 349         loff_t offset = iocb->ki_pos;
 350         struct inode *inode = file_inode(iocb->ki_filp);
 351
 352         if (error)
 353                 return error;
 354
 355         if (size && flags & IOMAP_DIO_UNWRITTEN)
 356                 return ext4_convert_unwritten_extents(NULL, inode,
 357                                                       offset, size);
 358
 359         return 0;
 360 }
 361
 362 static const struct iomap_dio_ops ext4_dio_write_ops = {
 363         .end_io = ext4_dio_write_end_io,
 364 };
 365
 366 static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
 367 {
 368         ssize_t ret;
 369         size_t count;
 370         loff_t offset;
 371         handle_t *handle;
 372         struct inode *inode = file_inode(iocb->ki_filp);
 373         bool extend = false, overwrite = false, unaligned_aio = false;
 374
 375         if (iocb->ki_flags & IOCB_NOWAIT) {
 376                 if (!inode_trylock(inode))
 377                         return -EAGAIN;
 378         } else {
 379                 inode_lock(inode);
 380         }
 381
 382         if (!ext4_dio_supported(inode)) {
 383                 inode_unlock(inode);
 384                 /*
 385                  * Fallback to buffered I/O if the inode does not support
 386                  * direct I/O.
 387                  */
 388                 return ext4_buffered_write_iter(iocb, from);
 389         }
 390
 391         ret = ext4_write_checks(iocb, from);
 392         if (ret <= 0) {
 393                 inode_unlock(inode);
 394                 return ret;
 395         }
 396
 397         /*
 398          * Unaligned asynchronous direct I/O must be serialized among each
 399          * other as the zeroing of partial blocks of two competing unaligned
 400          * asynchronous direct I/O writes can result in data corruption.
 401          */
 402         offset = iocb->ki_pos;
 403         count = iov_iter_count(from);
 404         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
 405             !is_sync_kiocb(iocb) && ext4_unaligned_aio(inode, from, offset)) {
 406                 unaligned_aio = true;
 407                 inode_dio_wait(inode);
 408         }
 409
 410         /*
 411          * Determine whether the I/O will overwrite allocated and initialized
 412          * blocks. If so, check to see whether it is possible to take the
 413          * dioread_nolock path.
 414          */
 415         if (!unaligned_aio && ext4_overwrite_io(inode, offset, count) &&
 416             ext4_should_dioread_nolock(inode)) {
 417                 overwrite = true;
 418                 downgrade_write(&inode->i_rwsem);
 419         }
 420
 421         if (offset + count > EXT4_I(inode)->i_disksize) {
 422                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
 423                 if (IS_ERR(handle)) {
 424                         ret = PTR_ERR(handle);
 425                         goto out;
 426                 }
 427
 428                 ret = ext4_orphan_add(handle, inode);
 429                 if (ret) {
 430                         ext4_journal_stop(handle);
 431                         goto out;
 432                 }
 433
 434                 extend = true;
 435                 ext4_journal_stop(handle);
 436         }
 437
 438         ret = iomap_dio_rw(iocb, from, &ext4_iomap_ops, &ext4_dio_write_ops,
 439                            is_sync_kiocb(iocb) || unaligned_aio || extend);
 440
 441         if (extend)
 442                 ret = ext4_handle_inode_extension(inode, offset, ret, count);
 443
 444 out:
 445         if (overwrite)
 446                 inode_unlock_shared(inode);
 447         else
 448                 inode_unlock(inode);
 449
 450         if (ret >= 0 && iov_iter_count(from)) {
 451                 ssize_t err;
 452                 loff_t endbyte;
 453
 454                 offset = iocb->ki_pos;
 455                 err = ext4_buffered_write_iter(iocb, from);
 456                 if (err < 0)
 457                         return err;
 458
 459                 /*
 460                  * We need to ensure that the pages within the page cache for
 461                  * the range covered by this I/O are written to disk and
 462                  * invalidated. This is in attempt to preserve the expected
 463                  * direct I/O semantics in the case we fallback to buffered I/O
 464                  * to complete off the I/O request.
 465                  */
 466                 ret += err;
 467                 endbyte = offset + err - 1;
 468                 err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
 469                                                    offset, endbyte);
 470                 if (!err)
 471                         invalidate_mapping_pages(iocb->ki_filp->f_mapping,
 472                                                  offset >> PAGE_SHIFT,
 473                                                  endbyte >> PAGE_SHIFT);
 474         }
 475
 476         return ret;
 477 }
 478
 479 #ifdef CONFIG_FS_DAX
 480 static ssize_t
 481 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
 482 {
 483         ssize_t ret;
 484         size_t count;
 485         loff_t offset;
 486         handle_t *handle;
 487         bool extend = false;
 488         struct inode *inode = file_inode(iocb->ki_filp);
 489
 490         if (!inode_trylock(inode)) {
 491                 if (iocb->ki_flags & IOCB_NOWAIT)
 492                         return -EAGAIN;
 493                 inode_lock(inode);
 494         }
 495
 496         ret = ext4_write_checks(iocb, from);
 497         if (ret <= 0)
 498                 goto out;
 499
 500         offset = iocb->ki_pos;
 501         count = iov_iter_count(from);
 502
 503         if (offset + count > EXT4_I(inode)->i_disksize) {
 504                 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
 505                 if (IS_ERR(handle)) {
 506                         ret = PTR_ERR(handle);
 507                         goto out;
 508                 }
 509
 510                 ret = ext4_orphan_add(handle, inode);
 511                 if (ret) {
 512                         ext4_journal_stop(handle);
 513                         goto out;
 514                 }
 515
 516                 extend = true;
 517                 ext4_journal_stop(handle);
 518         }
 519
 520         ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
 521
 522         if (extend)
 523                 ret = ext4_handle_inode_extension(inode, offset, ret, count);
 524 out:
 525         inode_unlock(inode);
 526         if (ret > 0)
 527                 ret = generic_write_sync(iocb, ret);
 528         return ret;
 529 }
 530 #endif
 531
 532 static ssize_t
 533 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 534 {
 535         struct inode *inode = file_inode(iocb->ki_filp);
 536
 537         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 538                 return -EIO;
 539
 540 #ifdef CONFIG_FS_DAX
 541         if (IS_DAX(inode))
 542                 return ext4_dax_write_iter(iocb, from);
 543 #endif
 544         if (iocb->ki_flags & IOCB_DIRECT)
 545                 return ext4_dio_write_iter(iocb, from);
 546
 547         return ext4_buffered_write_iter(iocb, from);
 548 }
 549
 550 #ifdef CONFIG_FS_DAX
 551 static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
 552                 enum page_entry_size pe_size)
 553 {
 554         int error = 0;
 555         vm_fault_t result;
 556         int retries = 0;
 557         handle_t *handle = NULL;
 558         struct inode *inode = file_inode(vmf->vma->vm_file);
 559         struct super_block *sb = inode->i_sb;
 560
 561         /*
 562          * We have to distinguish real writes from writes which will result in a
 563          * COW page; COW writes should *not* poke the journal (the file will not
 564          * be changed). Doing so would cause unintended failures when mounted
 565          * read-only.
 566          *
 567          * We check for VM_SHARED rather than vmf->cow_page since the latter is
 568          * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
 569          * other sizes, dax_iomap_fault will handle splitting / fallback so that
 570          * we eventually come back with a COW page.
 571          */
 572         bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
 573                 (vmf->vma->vm_flags & VM_SHARED);
 574         pfn_t pfn;
 575
 576         if (write) {
 577                 sb_start_pagefault(sb);
 578                 file_update_time(vmf->vma->vm_file);
 579                 down_read(&EXT4_I(inode)->i_mmap_sem);
 580 retry:
 581                 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
 582                                                EXT4_DATA_TRANS_BLOCKS(sb));
 583                 if (IS_ERR(handle)) {
 584                         up_read(&EXT4_I(inode)->i_mmap_sem);
 585                         sb_end_pagefault(sb);
 586                         return VM_FAULT_SIGBUS;
 587                 }
 588         } else {
 589                 down_read(&EXT4_I(inode)->i_mmap_sem);
 590         }
 591         result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
 592         if (write) {
 593                 ext4_journal_stop(handle);
 594
 595                 if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
 596                     ext4_should_retry_alloc(sb, &retries))
 597                         goto retry;
 598                 /* Handling synchronous page fault? */
 599                 if (result & VM_FAULT_NEEDDSYNC)
 600                         result = dax_finish_sync_fault(vmf, pe_size, pfn);
 601                 up_read(&EXT4_I(inode)->i_mmap_sem);
 602                 sb_end_pagefault(sb);
 603         } else {
 604                 up_read(&EXT4_I(inode)->i_mmap_sem);
 605         }
 606
 607         return result;
 608 }
 609
 610 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
 611 {
 612         return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
 613 }
 614
 615 static const struct vm_operations_struct ext4_dax_vm_ops = {
 616         .fault          = ext4_dax_fault,
 617         .huge_fault     = ext4_dax_huge_fault,
 618         .page_mkwrite   = ext4_dax_fault,
 619         .pfn_mkwrite    = ext4_dax_fault,
 620 };
 621 #else
 622 #define ext4_dax_vm_ops ext4_file_vm_ops
 623 #endif
 624
 625 static const struct vm_operations_struct ext4_file_vm_ops = {
 626         .fault          = ext4_filemap_fault,
 627         .map_pages      = filemap_map_pages,
 628         .page_mkwrite   = ext4_page_mkwrite,
 629 };
 630
 631 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
 632 {
 633         struct inode *inode = file->f_mapping->host;
 634         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 635         struct dax_device *dax_dev = sbi->s_daxdev;
 636
 637         if (unlikely(ext4_forced_shutdown(sbi)))
 638                 return -EIO;
 639
 640         /*
 641          * We don't support synchronous mappings for non-DAX files and
 642          * for DAX files if underneath dax_device is not synchronous.
 643          */
 644         if (!daxdev_mapping_supported(vma, dax_dev))
 645                 return -EOPNOTSUPP;
 646
 647         file_accessed(file);
 648         if (IS_DAX(file_inode(file))) {
 649                 vma->vm_ops = &ext4_dax_vm_ops;
 650                 vma->vm_flags |= VM_HUGEPAGE;
 651         } else {
 652                 vma->vm_ops = &ext4_file_vm_ops;
 653         }
 654         return 0;
 655 }
 656
 657 static int ext4_sample_last_mounted(struct super_block *sb,
 658                                     struct vfsmount *mnt)
 659 {
 660         struct ext4_sb_info *sbi = EXT4_SB(sb);
 661         struct path path;
 662         char buf[64], *cp;
 663         handle_t *handle;
 664         int err;
 665
 666         if (likely(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED))
 667                 return 0;
 668
 669         if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
 670                 return 0;
 671
 672         sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
 673         /*
 674          * Sample where the filesystem has been mounted and
 675          * store it in the superblock for sysadmin convenience
 676          * when trying to sort through large numbers of block
 677          * devices or filesystem images.
 678          */
 679         memset(buf, 0, sizeof(buf));
 680         path.mnt = mnt;
 681         path.dentry = mnt->mnt_root;
 682         cp = d_path(&path, buf, sizeof(buf));
 683         err = 0;
 684         if (IS_ERR(cp))
 685                 goto out;
 686
 687         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
 688         err = PTR_ERR(handle);
 689         if (IS_ERR(handle))
 690                 goto out;
 691         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
 692         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
 693         if (err)
 694                 goto out_journal;
 695         strlcpy(sbi->s_es->s_last_mounted, cp,
 696                 sizeof(sbi->s_es->s_last_mounted));
 697         ext4_handle_dirty_super(handle, sb);
 698 out_journal:
 699         ext4_journal_stop(handle);
 700 out:
 701         sb_end_intwrite(sb);
 702         return err;
 703 }
 704
 705 static int ext4_file_open(struct inode * inode, struct file * filp)
 706 {
 707         int ret;
 708
 709         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 710                 return -EIO;
 711
 712         ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
 713         if (ret)
 714                 return ret;
 715
 716         ret = fscrypt_file_open(inode, filp);
 717         if (ret)
 718                 return ret;
 719
 720         ret = fsverity_file_open(inode, filp);
 721         if (ret)
 722                 return ret;
 723
 724         /*
 725          * Set up the jbd2_inode if we are opening the inode for
 726          * writing and the journal is present
 727          */
 728         if (filp->f_mode & FMODE_WRITE) {
 729                 ret = ext4_inode_attach_jinode(inode);
 730                 if (ret < 0)
 731                         return ret;
 732         }
 733
 734         filp->f_mode |= FMODE_NOWAIT;
 735         return dquot_file_open(inode, filp);
 736 }
 737
 738 /*
 739  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
 740  * by calling generic_file_llseek_size() with the appropriate maxbytes
 741  * value for each.
 742  */
 743 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
 744 {
 745         struct inode *inode = file->f_mapping->host;
 746         loff_t maxbytes;
 747
 748         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 749                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
 750         else
 751                 maxbytes = inode->i_sb->s_maxbytes;
 752
 753         switch (whence) {
 754         default:
 755                 return generic_file_llseek_size(file, offset, whence,
 756                                                 maxbytes, i_size_read(inode));
 757         case SEEK_HOLE:
 758                 inode_lock_shared(inode);
 759                 offset = iomap_seek_hole(inode, offset,
 760                                          &ext4_iomap_report_ops);
 761                 inode_unlock_shared(inode);
 762                 break;
 763         case SEEK_DATA:
 764                 inode_lock_shared(inode);
 765                 offset = iomap_seek_data(inode, offset,
 766                                          &ext4_iomap_report_ops);
 767                 inode_unlock_shared(inode);
 768                 break;
 769         }
 770
 771         if (offset < 0)
 772                 return offset;
 773         return vfs_setpos(file, offset, maxbytes);
 774 }
 775
 776 const struct file_operations ext4_file_operations = {
 777         .llseek         = ext4_llseek,
 778         .read_iter      = ext4_file_read_iter,
 779         .write_iter     = ext4_file_write_iter,
 780         .unlocked_ioctl = ext4_ioctl,
 781 #ifdef CONFIG_COMPAT
 782         .compat_ioctl   = ext4_compat_ioctl,
 783 #endif
 784         .mmap           = ext4_file_mmap,
 785         .mmap_supported_flags = MAP_SYNC,
 786         .open           = ext4_file_open,
 787         .release        = ext4_release_file,
 788         .fsync          = ext4_sync_file,
 789         .get_unmapped_area = thp_get_unmapped_area,
 790         .splice_read    = generic_file_splice_read,
 791         .splice_write   = iter_file_splice_write,
 792         .fallocate      = ext4_fallocate,
 793 };
 794
 795 const struct inode_operations ext4_file_inode_operations = {
 796         .setattr        = ext4_setattr,
 797         .getattr        = ext4_file_getattr,
 798         .listxattr      = ext4_listxattr,
 799         .get_acl        = ext4_get_acl,
 800         .set_acl        = ext4_set_acl,
 801         .fiemap         = ext4_fiemap,
 802 };
 803