1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or https://opensource.org/licenses/CDDL-1.0. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2011, Lawrence Livermore National Security, LLC. 23 * Copyright (c) 2015 by Chunwei Chen. All rights reserved. 24 */ 25 26 27 #ifdef CONFIG_COMPAT 28 #include <linux/compat.h> 29 #endif 30 #include <linux/fs.h> 31 #include <sys/file.h> 32 #include <sys/dmu_objset.h> 33 #include <sys/zfs_znode.h> 34 #include <sys/zfs_vfsops.h> 35 #include <sys/zfs_vnops.h> 36 #include <sys/zfs_project.h> 37 #if defined(HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS) || \ 38 defined(HAVE_VFS_FILEMAP_DIRTY_FOLIO) 39 #include <linux/pagemap.h> 40 #endif 41 #ifdef HAVE_FILE_FADVISE 42 #include <linux/fadvise.h> 43 #endif 44 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO 45 #include <linux/writeback.h> 46 #endif 47 48 /* 49 * When using fallocate(2) to preallocate space, inflate the requested 50 * capacity check by 10% to account for the required metadata blocks. 51 */ 52 static unsigned int zfs_fallocate_reserve_percent = 110; 53 54 static int 55 zpl_open(struct inode *ip, struct file *filp) 56 { 57 cred_t *cr = CRED(); 58 int error; 59 fstrans_cookie_t cookie; 60 61 error = generic_file_open(ip, filp); 62 if (error) 63 return (error); 64 65 crhold(cr); 66 cookie = spl_fstrans_mark(); 67 error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr); 68 spl_fstrans_unmark(cookie); 69 crfree(cr); 70 ASSERT3S(error, <=, 0); 71 72 return (error); 73 } 74 75 static int 76 zpl_release(struct inode *ip, struct file *filp) 77 { 78 cred_t *cr = CRED(); 79 int error; 80 fstrans_cookie_t cookie; 81 82 cookie = spl_fstrans_mark(); 83 if (ITOZ(ip)->z_atime_dirty) 84 zfs_mark_inode_dirty(ip); 85 86 crhold(cr); 87 error = -zfs_close(ip, filp->f_flags, cr); 88 spl_fstrans_unmark(cookie); 89 crfree(cr); 90 ASSERT3S(error, <=, 0); 91 92 return (error); 93 } 94 95 static int 96 zpl_iterate(struct file *filp, zpl_dir_context_t *ctx) 97 { 98 cred_t *cr = CRED(); 99 int error; 100 fstrans_cookie_t cookie; 101 102 crhold(cr); 103 cookie = spl_fstrans_mark(); 104 error = -zfs_readdir(file_inode(filp), ctx, cr); 105 spl_fstrans_unmark(cookie); 106 crfree(cr); 107 ASSERT3S(error, <=, 0); 108 109 return (error); 110 } 111 112 #if !defined(HAVE_VFS_ITERATE) && !defined(HAVE_VFS_ITERATE_SHARED) 113 static int 114 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir) 115 { 116 zpl_dir_context_t ctx = 117 ZPL_DIR_CONTEXT_INIT(dirent, filldir, filp->f_pos); 118 int error; 119 120 error = zpl_iterate(filp, &ctx); 121 filp->f_pos = ctx.pos; 122 123 return (error); 124 } 125 #endif /* !HAVE_VFS_ITERATE && !HAVE_VFS_ITERATE_SHARED */ 126 127 #if defined(HAVE_FSYNC_WITHOUT_DENTRY) 128 /* 129 * Linux 2.6.35 - 3.0 API, 130 * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed 131 * redundant. The dentry is still accessible via filp->f_path.dentry, 132 * and we are guaranteed that filp will never be NULL. 133 */ 134 static int 135 zpl_fsync(struct file *filp, int datasync) 136 { 137 struct inode *inode = filp->f_mapping->host; 138 cred_t *cr = CRED(); 139 int error; 140 fstrans_cookie_t cookie; 141 142 crhold(cr); 143 cookie = spl_fstrans_mark(); 144 error = -zfs_fsync(ITOZ(inode), datasync, cr); 145 spl_fstrans_unmark(cookie); 146 crfree(cr); 147 ASSERT3S(error, <=, 0); 148 149 return (error); 150 } 151 152 #ifdef HAVE_FILE_AIO_FSYNC 153 static int 154 zpl_aio_fsync(struct kiocb *kiocb, int datasync) 155 { 156 return (zpl_fsync(kiocb->ki_filp, datasync)); 157 } 158 #endif 159 160 #elif defined(HAVE_FSYNC_RANGE) 161 /* 162 * Linux 3.1 API, 163 * As of 3.1 the responsibility to call filemap_write_and_wait_range() has 164 * been pushed down in to the .fsync() vfs hook. Additionally, the i_mutex 165 * lock is no longer held by the caller, for zfs we don't require the lock 166 * to be held so we don't acquire it. 167 */ 168 static int 169 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync) 170 { 171 struct inode *inode = filp->f_mapping->host; 172 znode_t *zp = ITOZ(inode); 173 zfsvfs_t *zfsvfs = ITOZSB(inode); 174 cred_t *cr = CRED(); 175 int error; 176 fstrans_cookie_t cookie; 177 178 /* 179 * The variables z_sync_writes_cnt and z_async_writes_cnt work in 180 * tandem so that sync writes can detect if there are any non-sync 181 * writes going on and vice-versa. The "vice-versa" part to this logic 182 * is located in zfs_putpage() where non-sync writes check if there are 183 * any ongoing sync writes. If any sync and non-sync writes overlap, 184 * we do a commit to complete the non-sync writes since the latter can 185 * potentially take several seconds to complete and thus block sync 186 * writes in the upcoming call to filemap_write_and_wait_range(). 187 */ 188 atomic_inc_32(&zp->z_sync_writes_cnt); 189 /* 190 * If the following check does not detect an overlapping non-sync write 191 * (say because it's just about to start), then it is guaranteed that 192 * the non-sync write will detect this sync write. This is because we 193 * always increment z_sync_writes_cnt / z_async_writes_cnt before doing 194 * the check on z_async_writes_cnt / z_sync_writes_cnt here and in 195 * zfs_putpage() respectively. 196 */ 197 if (atomic_load_32(&zp->z_async_writes_cnt) > 0) { 198 if ((error = zpl_enter(zfsvfs, FTAG)) != 0) { 199 atomic_dec_32(&zp->z_sync_writes_cnt); 200 return (error); 201 } 202 zil_commit(zfsvfs->z_log, zp->z_id); 203 zpl_exit(zfsvfs, FTAG); 204 } 205 206 error = filemap_write_and_wait_range(inode->i_mapping, start, end); 207 208 /* 209 * The sync write is not complete yet but we decrement 210 * z_sync_writes_cnt since zfs_fsync() increments and decrements 211 * it internally. If a non-sync write starts just after the decrement 212 * operation but before we call zfs_fsync(), it may not detect this 213 * overlapping sync write but it does not matter since we have already 214 * gone past filemap_write_and_wait_range() and we won't block due to 215 * the non-sync write. 216 */ 217 atomic_dec_32(&zp->z_sync_writes_cnt); 218 219 if (error) 220 return (error); 221 222 crhold(cr); 223 cookie = spl_fstrans_mark(); 224 error = -zfs_fsync(zp, datasync, cr); 225 spl_fstrans_unmark(cookie); 226 crfree(cr); 227 ASSERT3S(error, <=, 0); 228 229 return (error); 230 } 231 232 #ifdef HAVE_FILE_AIO_FSYNC 233 static int 234 zpl_aio_fsync(struct kiocb *kiocb, int datasync) 235 { 236 return (zpl_fsync(kiocb->ki_filp, kiocb->ki_pos, -1, datasync)); 237 } 238 #endif 239 240 #else 241 #error "Unsupported fops->fsync() implementation" 242 #endif 243 244 static inline int 245 zfs_io_flags(struct kiocb *kiocb) 246 { 247 int flags = 0; 248 249 #if defined(IOCB_DSYNC) 250 if (kiocb->ki_flags & IOCB_DSYNC) 251 flags |= O_DSYNC; 252 #endif 253 #if defined(IOCB_SYNC) 254 if (kiocb->ki_flags & IOCB_SYNC) 255 flags |= O_SYNC; 256 #endif 257 #if defined(IOCB_APPEND) 258 if (kiocb->ki_flags & IOCB_APPEND) 259 flags |= O_APPEND; 260 #endif 261 #if defined(IOCB_DIRECT) 262 if (kiocb->ki_flags & IOCB_DIRECT) 263 flags |= O_DIRECT; 264 #endif 265 return (flags); 266 } 267 268 /* 269 * If relatime is enabled, call file_accessed() if zfs_relatime_need_update() 270 * is true. This is needed since datasets with inherited "relatime" property 271 * aren't necessarily mounted with the MNT_RELATIME flag (e.g. after 272 * `zfs set relatime=...`), which is what relatime test in VFS by 273 * relatime_need_update() is based on. 274 */ 275 static inline void 276 zpl_file_accessed(struct file *filp) 277 { 278 struct inode *ip = filp->f_mapping->host; 279 280 if (!IS_NOATIME(ip) && ITOZSB(ip)->z_relatime) { 281 if (zfs_relatime_need_update(ip)) 282 file_accessed(filp); 283 } else { 284 file_accessed(filp); 285 } 286 } 287 288 #if defined(HAVE_VFS_RW_ITERATE) 289 290 /* 291 * When HAVE_VFS_IOV_ITER is defined the iov_iter structure supports 292 * iovecs, kvevs, bvecs and pipes, plus all the required interfaces to 293 * manipulate the iov_iter are available. In which case the full iov_iter 294 * can be attached to the uio and correctly handled in the lower layers. 295 * Otherwise, for older kernels extract the iovec and pass it instead. 296 */ 297 static void 298 zpl_uio_init(zfs_uio_t *uio, struct kiocb *kiocb, struct iov_iter *to, 299 loff_t pos, ssize_t count, size_t skip) 300 { 301 #if defined(HAVE_VFS_IOV_ITER) 302 zfs_uio_iov_iter_init(uio, to, pos, count, skip); 303 #else 304 zfs_uio_iovec_init(uio, zfs_uio_iter_iov(to), to->nr_segs, pos, 305 zfs_uio_iov_iter_type(to) & ITER_KVEC ? 306 UIO_SYSSPACE : UIO_USERSPACE, 307 count, skip); 308 #endif 309 } 310 311 static ssize_t 312 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to) 313 { 314 cred_t *cr = CRED(); 315 fstrans_cookie_t cookie; 316 struct file *filp = kiocb->ki_filp; 317 ssize_t count = iov_iter_count(to); 318 zfs_uio_t uio; 319 320 zpl_uio_init(&uio, kiocb, to, kiocb->ki_pos, count, 0); 321 322 crhold(cr); 323 cookie = spl_fstrans_mark(); 324 325 int error = -zfs_read(ITOZ(filp->f_mapping->host), &uio, 326 filp->f_flags | zfs_io_flags(kiocb), cr); 327 328 spl_fstrans_unmark(cookie); 329 crfree(cr); 330 331 if (error < 0) 332 return (error); 333 334 ssize_t read = count - uio.uio_resid; 335 kiocb->ki_pos += read; 336 337 zpl_file_accessed(filp); 338 339 return (read); 340 } 341 342 static inline ssize_t 343 zpl_generic_write_checks(struct kiocb *kiocb, struct iov_iter *from, 344 size_t *countp) 345 { 346 #ifdef HAVE_GENERIC_WRITE_CHECKS_KIOCB 347 ssize_t ret = generic_write_checks(kiocb, from); 348 if (ret <= 0) 349 return (ret); 350 351 *countp = ret; 352 #else 353 struct file *file = kiocb->ki_filp; 354 struct address_space *mapping = file->f_mapping; 355 struct inode *ip = mapping->host; 356 int isblk = S_ISBLK(ip->i_mode); 357 358 *countp = iov_iter_count(from); 359 ssize_t ret = generic_write_checks(file, &kiocb->ki_pos, countp, isblk); 360 if (ret) 361 return (ret); 362 #endif 363 364 return (0); 365 } 366 367 static ssize_t 368 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from) 369 { 370 cred_t *cr = CRED(); 371 fstrans_cookie_t cookie; 372 struct file *filp = kiocb->ki_filp; 373 struct inode *ip = filp->f_mapping->host; 374 zfs_uio_t uio; 375 size_t count = 0; 376 ssize_t ret; 377 378 ret = zpl_generic_write_checks(kiocb, from, &count); 379 if (ret) 380 return (ret); 381 382 zpl_uio_init(&uio, kiocb, from, kiocb->ki_pos, count, from->iov_offset); 383 384 crhold(cr); 385 cookie = spl_fstrans_mark(); 386 387 int error = -zfs_write(ITOZ(ip), &uio, 388 filp->f_flags | zfs_io_flags(kiocb), cr); 389 390 spl_fstrans_unmark(cookie); 391 crfree(cr); 392 393 if (error < 0) 394 return (error); 395 396 ssize_t wrote = count - uio.uio_resid; 397 kiocb->ki_pos += wrote; 398 399 return (wrote); 400 } 401 402 #else /* !HAVE_VFS_RW_ITERATE */ 403 404 static ssize_t 405 zpl_aio_read(struct kiocb *kiocb, const struct iovec *iov, 406 unsigned long nr_segs, loff_t pos) 407 { 408 cred_t *cr = CRED(); 409 fstrans_cookie_t cookie; 410 struct file *filp = kiocb->ki_filp; 411 size_t count; 412 ssize_t ret; 413 414 ret = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 415 if (ret) 416 return (ret); 417 418 zfs_uio_t uio; 419 zfs_uio_iovec_init(&uio, iov, nr_segs, kiocb->ki_pos, UIO_USERSPACE, 420 count, 0); 421 422 crhold(cr); 423 cookie = spl_fstrans_mark(); 424 425 int error = -zfs_read(ITOZ(filp->f_mapping->host), &uio, 426 filp->f_flags | zfs_io_flags(kiocb), cr); 427 428 spl_fstrans_unmark(cookie); 429 crfree(cr); 430 431 if (error < 0) 432 return (error); 433 434 ssize_t read = count - uio.uio_resid; 435 kiocb->ki_pos += read; 436 437 zpl_file_accessed(filp); 438 439 return (read); 440 } 441 442 static ssize_t 443 zpl_aio_write(struct kiocb *kiocb, const struct iovec *iov, 444 unsigned long nr_segs, loff_t pos) 445 { 446 cred_t *cr = CRED(); 447 fstrans_cookie_t cookie; 448 struct file *filp = kiocb->ki_filp; 449 struct inode *ip = filp->f_mapping->host; 450 size_t count; 451 ssize_t ret; 452 453 ret = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); 454 if (ret) 455 return (ret); 456 457 ret = generic_write_checks(filp, &pos, &count, S_ISBLK(ip->i_mode)); 458 if (ret) 459 return (ret); 460 461 kiocb->ki_pos = pos; 462 463 zfs_uio_t uio; 464 zfs_uio_iovec_init(&uio, iov, nr_segs, kiocb->ki_pos, UIO_USERSPACE, 465 count, 0); 466 467 crhold(cr); 468 cookie = spl_fstrans_mark(); 469 470 int error = -zfs_write(ITOZ(ip), &uio, 471 filp->f_flags | zfs_io_flags(kiocb), cr); 472 473 spl_fstrans_unmark(cookie); 474 crfree(cr); 475 476 if (error < 0) 477 return (error); 478 479 ssize_t wrote = count - uio.uio_resid; 480 kiocb->ki_pos += wrote; 481 482 return (wrote); 483 } 484 #endif /* HAVE_VFS_RW_ITERATE */ 485 486 #if defined(HAVE_VFS_RW_ITERATE) 487 static ssize_t 488 zpl_direct_IO_impl(int rw, struct kiocb *kiocb, struct iov_iter *iter) 489 { 490 if (rw == WRITE) 491 return (zpl_iter_write(kiocb, iter)); 492 else 493 return (zpl_iter_read(kiocb, iter)); 494 } 495 #if defined(HAVE_VFS_DIRECT_IO_ITER) 496 static ssize_t 497 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter) 498 { 499 return (zpl_direct_IO_impl(iov_iter_rw(iter), kiocb, iter)); 500 } 501 #elif defined(HAVE_VFS_DIRECT_IO_ITER_OFFSET) 502 static ssize_t 503 zpl_direct_IO(struct kiocb *kiocb, struct iov_iter *iter, loff_t pos) 504 { 505 ASSERT3S(pos, ==, kiocb->ki_pos); 506 return (zpl_direct_IO_impl(iov_iter_rw(iter), kiocb, iter)); 507 } 508 #elif defined(HAVE_VFS_DIRECT_IO_ITER_RW_OFFSET) 509 static ssize_t 510 zpl_direct_IO(int rw, struct kiocb *kiocb, struct iov_iter *iter, loff_t pos) 511 { 512 ASSERT3S(pos, ==, kiocb->ki_pos); 513 return (zpl_direct_IO_impl(rw, kiocb, iter)); 514 } 515 #else 516 #error "Unknown direct IO interface" 517 #endif 518 519 #else /* HAVE_VFS_RW_ITERATE */ 520 521 #if defined(HAVE_VFS_DIRECT_IO_IOVEC) 522 static ssize_t 523 zpl_direct_IO(int rw, struct kiocb *kiocb, const struct iovec *iov, 524 loff_t pos, unsigned long nr_segs) 525 { 526 if (rw == WRITE) 527 return (zpl_aio_write(kiocb, iov, nr_segs, pos)); 528 else 529 return (zpl_aio_read(kiocb, iov, nr_segs, pos)); 530 } 531 #elif defined(HAVE_VFS_DIRECT_IO_ITER_RW_OFFSET) 532 static ssize_t 533 zpl_direct_IO(int rw, struct kiocb *kiocb, struct iov_iter *iter, loff_t pos) 534 { 535 const struct iovec *iovp = iov_iter_iovec(iter); 536 unsigned long nr_segs = iter->nr_segs; 537 538 ASSERT3S(pos, ==, kiocb->ki_pos); 539 if (rw == WRITE) 540 return (zpl_aio_write(kiocb, iovp, nr_segs, pos)); 541 else 542 return (zpl_aio_read(kiocb, iovp, nr_segs, pos)); 543 } 544 #else 545 #error "Unknown direct IO interface" 546 #endif 547 548 #endif /* HAVE_VFS_RW_ITERATE */ 549 550 static loff_t 551 zpl_llseek(struct file *filp, loff_t offset, int whence) 552 { 553 #if defined(SEEK_HOLE) && defined(SEEK_DATA) 554 fstrans_cookie_t cookie; 555 556 if (whence == SEEK_DATA || whence == SEEK_HOLE) { 557 struct inode *ip = filp->f_mapping->host; 558 loff_t maxbytes = ip->i_sb->s_maxbytes; 559 loff_t error; 560 561 spl_inode_lock_shared(ip); 562 cookie = spl_fstrans_mark(); 563 error = -zfs_holey(ITOZ(ip), whence, &offset); 564 spl_fstrans_unmark(cookie); 565 if (error == 0) 566 error = lseek_execute(filp, ip, offset, maxbytes); 567 spl_inode_unlock_shared(ip); 568 569 return (error); 570 } 571 #endif /* SEEK_HOLE && SEEK_DATA */ 572 573 return (generic_file_llseek(filp, offset, whence)); 574 } 575 576 /* 577 * It's worth taking a moment to describe how mmap is implemented 578 * for zfs because it differs considerably from other Linux filesystems. 579 * However, this issue is handled the same way under OpenSolaris. 580 * 581 * The issue is that by design zfs bypasses the Linux page cache and 582 * leaves all caching up to the ARC. This has been shown to work 583 * well for the common read(2)/write(2) case. However, mmap(2) 584 * is problem because it relies on being tightly integrated with the 585 * page cache. To handle this we cache mmap'ed files twice, once in 586 * the ARC and a second time in the page cache. The code is careful 587 * to keep both copies synchronized. 588 * 589 * When a file with an mmap'ed region is written to using write(2) 590 * both the data in the ARC and existing pages in the page cache 591 * are updated. For a read(2) data will be read first from the page 592 * cache then the ARC if needed. Neither a write(2) or read(2) will 593 * will ever result in new pages being added to the page cache. 594 * 595 * New pages are added to the page cache only via .readpage() which 596 * is called when the vfs needs to read a page off disk to back the 597 * virtual memory region. These pages may be modified without 598 * notifying the ARC and will be written out periodically via 599 * .writepage(). This will occur due to either a sync or the usual 600 * page aging behavior. Note because a read(2) of a mmap'ed file 601 * will always check the page cache first even when the ARC is out 602 * of date correct data will still be returned. 603 * 604 * While this implementation ensures correct behavior it does have 605 * have some drawbacks. The most obvious of which is that it 606 * increases the required memory footprint when access mmap'ed 607 * files. It also adds additional complexity to the code keeping 608 * both caches synchronized. 609 * 610 * Longer term it may be possible to cleanly resolve this wart by 611 * mapping page cache pages directly on to the ARC buffers. The 612 * Linux address space operations are flexible enough to allow 613 * selection of which pages back a particular index. The trick 614 * would be working out the details of which subsystem is in 615 * charge, the ARC, the page cache, or both. It may also prove 616 * helpful to move the ARC buffers to a scatter-gather lists 617 * rather than a vmalloc'ed region. 618 */ 619 static int 620 zpl_mmap(struct file *filp, struct vm_area_struct *vma) 621 { 622 struct inode *ip = filp->f_mapping->host; 623 int error; 624 fstrans_cookie_t cookie; 625 626 cookie = spl_fstrans_mark(); 627 error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start, 628 (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags); 629 spl_fstrans_unmark(cookie); 630 if (error) 631 return (error); 632 633 error = generic_file_mmap(filp, vma); 634 if (error) 635 return (error); 636 637 #if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE) 638 znode_t *zp = ITOZ(ip); 639 mutex_enter(&zp->z_lock); 640 zp->z_is_mapped = B_TRUE; 641 mutex_exit(&zp->z_lock); 642 #endif 643 644 return (error); 645 } 646 647 /* 648 * Populate a page with data for the Linux page cache. This function is 649 * only used to support mmap(2). There will be an identical copy of the 650 * data in the ARC which is kept up to date via .write() and .writepage(). 651 */ 652 static inline int 653 zpl_readpage_common(struct page *pp) 654 { 655 fstrans_cookie_t cookie; 656 657 ASSERT(PageLocked(pp)); 658 659 cookie = spl_fstrans_mark(); 660 int error = -zfs_getpage(pp->mapping->host, pp); 661 spl_fstrans_unmark(cookie); 662 663 unlock_page(pp); 664 665 return (error); 666 } 667 668 #ifdef HAVE_VFS_READ_FOLIO 669 static int 670 zpl_read_folio(struct file *filp, struct folio *folio) 671 { 672 return (zpl_readpage_common(&folio->page)); 673 } 674 #else 675 static int 676 zpl_readpage(struct file *filp, struct page *pp) 677 { 678 return (zpl_readpage_common(pp)); 679 } 680 #endif 681 682 static int 683 zpl_readpage_filler(void *data, struct page *pp) 684 { 685 return (zpl_readpage_common(pp)); 686 } 687 688 /* 689 * Populate a set of pages with data for the Linux page cache. This 690 * function will only be called for read ahead and never for demand 691 * paging. For simplicity, the code relies on read_cache_pages() to 692 * correctly lock each page for IO and call zpl_readpage(). 693 */ 694 #ifdef HAVE_VFS_READPAGES 695 static int 696 zpl_readpages(struct file *filp, struct address_space *mapping, 697 struct list_head *pages, unsigned nr_pages) 698 { 699 return (read_cache_pages(mapping, pages, zpl_readpage_filler, NULL)); 700 } 701 #else 702 static void 703 zpl_readahead(struct readahead_control *ractl) 704 { 705 struct page *page; 706 707 while ((page = readahead_page(ractl)) != NULL) { 708 int ret; 709 710 ret = zpl_readpage_filler(NULL, page); 711 put_page(page); 712 if (ret) 713 break; 714 } 715 } 716 #endif 717 718 static int 719 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data) 720 { 721 boolean_t *for_sync = data; 722 fstrans_cookie_t cookie; 723 724 ASSERT(PageLocked(pp)); 725 ASSERT(!PageWriteback(pp)); 726 727 cookie = spl_fstrans_mark(); 728 (void) zfs_putpage(pp->mapping->host, pp, wbc, *for_sync); 729 spl_fstrans_unmark(cookie); 730 731 return (0); 732 } 733 734 #ifdef HAVE_WRITEPAGE_T_FOLIO 735 static int 736 zpl_putfolio(struct folio *pp, struct writeback_control *wbc, void *data) 737 { 738 (void) zpl_putpage(&pp->page, wbc, data); 739 return (0); 740 } 741 #endif 742 743 static inline int 744 zpl_write_cache_pages(struct address_space *mapping, 745 struct writeback_control *wbc, void *data) 746 { 747 int result; 748 749 #ifdef HAVE_WRITEPAGE_T_FOLIO 750 result = write_cache_pages(mapping, wbc, zpl_putfolio, data); 751 #else 752 result = write_cache_pages(mapping, wbc, zpl_putpage, data); 753 #endif 754 return (result); 755 } 756 757 static int 758 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc) 759 { 760 znode_t *zp = ITOZ(mapping->host); 761 zfsvfs_t *zfsvfs = ITOZSB(mapping->host); 762 enum writeback_sync_modes sync_mode; 763 int result; 764 765 if ((result = zpl_enter(zfsvfs, FTAG)) != 0) 766 return (result); 767 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 768 wbc->sync_mode = WB_SYNC_ALL; 769 zpl_exit(zfsvfs, FTAG); 770 sync_mode = wbc->sync_mode; 771 772 /* 773 * We don't want to run write_cache_pages() in SYNC mode here, because 774 * that would make putpage() wait for a single page to be committed to 775 * disk every single time, resulting in atrocious performance. Instead 776 * we run it once in non-SYNC mode so that the ZIL gets all the data, 777 * and then we commit it all in one go. 778 */ 779 boolean_t for_sync = (sync_mode == WB_SYNC_ALL); 780 wbc->sync_mode = WB_SYNC_NONE; 781 result = zpl_write_cache_pages(mapping, wbc, &for_sync); 782 if (sync_mode != wbc->sync_mode) { 783 if ((result = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 784 return (result); 785 if (zfsvfs->z_log != NULL) 786 zil_commit(zfsvfs->z_log, zp->z_id); 787 zpl_exit(zfsvfs, FTAG); 788 789 /* 790 * We need to call write_cache_pages() again (we can't just 791 * return after the commit) because the previous call in 792 * non-SYNC mode does not guarantee that we got all the dirty 793 * pages (see the implementation of write_cache_pages() for 794 * details). That being said, this is a no-op in most cases. 795 */ 796 wbc->sync_mode = sync_mode; 797 result = zpl_write_cache_pages(mapping, wbc, &for_sync); 798 } 799 return (result); 800 } 801 802 /* 803 * Write out dirty pages to the ARC, this function is only required to 804 * support mmap(2). Mapped pages may be dirtied by memory operations 805 * which never call .write(). These dirty pages are kept in sync with 806 * the ARC buffers via this hook. 807 */ 808 static int 809 zpl_writepage(struct page *pp, struct writeback_control *wbc) 810 { 811 if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS) 812 wbc->sync_mode = WB_SYNC_ALL; 813 814 boolean_t for_sync = (wbc->sync_mode == WB_SYNC_ALL); 815 816 return (zpl_putpage(pp, wbc, &for_sync)); 817 } 818 819 /* 820 * The flag combination which matches the behavior of zfs_space() is 821 * FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE. The FALLOC_FL_PUNCH_HOLE 822 * flag was introduced in the 2.6.38 kernel. 823 * 824 * The original mode=0 (allocate space) behavior can be reasonably emulated 825 * by checking if enough space exists and creating a sparse file, as real 826 * persistent space reservation is not possible due to COW, snapshots, etc. 827 */ 828 static long 829 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len) 830 { 831 cred_t *cr = CRED(); 832 loff_t olen; 833 fstrans_cookie_t cookie; 834 int error = 0; 835 836 int test_mode = FALLOC_FL_PUNCH_HOLE; 837 #ifdef HAVE_FALLOC_FL_ZERO_RANGE 838 test_mode |= FALLOC_FL_ZERO_RANGE; 839 #endif 840 841 if ((mode & ~(FALLOC_FL_KEEP_SIZE | test_mode)) != 0) 842 return (-EOPNOTSUPP); 843 844 if (offset < 0 || len <= 0) 845 return (-EINVAL); 846 847 spl_inode_lock(ip); 848 olen = i_size_read(ip); 849 850 crhold(cr); 851 cookie = spl_fstrans_mark(); 852 if (mode & (test_mode)) { 853 flock64_t bf; 854 855 if (mode & FALLOC_FL_KEEP_SIZE) { 856 if (offset > olen) 857 goto out_unmark; 858 859 if (offset + len > olen) 860 len = olen - offset; 861 } 862 bf.l_type = F_WRLCK; 863 bf.l_whence = SEEK_SET; 864 bf.l_start = offset; 865 bf.l_len = len; 866 bf.l_pid = 0; 867 868 error = -zfs_space(ITOZ(ip), F_FREESP, &bf, O_RDWR, offset, cr); 869 } else if ((mode & ~FALLOC_FL_KEEP_SIZE) == 0) { 870 unsigned int percent = zfs_fallocate_reserve_percent; 871 struct kstatfs statfs; 872 873 /* Legacy mode, disable fallocate compatibility. */ 874 if (percent == 0) { 875 error = -EOPNOTSUPP; 876 goto out_unmark; 877 } 878 879 /* 880 * Use zfs_statvfs() instead of dmu_objset_space() since it 881 * also checks project quota limits, which are relevant here. 882 */ 883 error = zfs_statvfs(ip, &statfs); 884 if (error) 885 goto out_unmark; 886 887 /* 888 * Shrink available space a bit to account for overhead/races. 889 * We know the product previously fit into availbytes from 890 * dmu_objset_space(), so the smaller product will also fit. 891 */ 892 if (len > statfs.f_bavail * (statfs.f_bsize * 100 / percent)) { 893 error = -ENOSPC; 894 goto out_unmark; 895 } 896 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > olen) 897 error = zfs_freesp(ITOZ(ip), offset + len, 0, 0, FALSE); 898 } 899 out_unmark: 900 spl_fstrans_unmark(cookie); 901 spl_inode_unlock(ip); 902 903 crfree(cr); 904 905 return (error); 906 } 907 908 static long 909 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len) 910 { 911 return zpl_fallocate_common(file_inode(filp), 912 mode, offset, len); 913 } 914 915 static int 916 zpl_ioctl_getversion(struct file *filp, void __user *arg) 917 { 918 uint32_t generation = file_inode(filp)->i_generation; 919 920 return (copy_to_user(arg, &generation, sizeof (generation))); 921 } 922 923 #ifdef HAVE_FILE_FADVISE 924 static int 925 zpl_fadvise(struct file *filp, loff_t offset, loff_t len, int advice) 926 { 927 struct inode *ip = file_inode(filp); 928 znode_t *zp = ITOZ(ip); 929 zfsvfs_t *zfsvfs = ITOZSB(ip); 930 objset_t *os = zfsvfs->z_os; 931 int error = 0; 932 933 if (S_ISFIFO(ip->i_mode)) 934 return (-ESPIPE); 935 936 if (offset < 0 || len < 0) 937 return (-EINVAL); 938 939 if ((error = zpl_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 940 return (error); 941 942 switch (advice) { 943 case POSIX_FADV_SEQUENTIAL: 944 case POSIX_FADV_WILLNEED: 945 #ifdef HAVE_GENERIC_FADVISE 946 if (zn_has_cached_data(zp, offset, offset + len - 1)) 947 error = generic_fadvise(filp, offset, len, advice); 948 #endif 949 /* 950 * Pass on the caller's size directly, but note that 951 * dmu_prefetch_max will effectively cap it. If there 952 * really is a larger sequential access pattern, perhaps 953 * dmu_zfetch will detect it. 954 */ 955 if (len == 0) 956 len = i_size_read(ip) - offset; 957 958 dmu_prefetch(os, zp->z_id, 0, offset, len, 959 ZIO_PRIORITY_ASYNC_READ); 960 break; 961 case POSIX_FADV_NORMAL: 962 case POSIX_FADV_RANDOM: 963 case POSIX_FADV_DONTNEED: 964 case POSIX_FADV_NOREUSE: 965 /* ignored for now */ 966 break; 967 default: 968 error = -EINVAL; 969 break; 970 } 971 972 zfs_exit(zfsvfs, FTAG); 973 974 return (error); 975 } 976 #endif /* HAVE_FILE_FADVISE */ 977 978 #define ZFS_FL_USER_VISIBLE (FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL) 979 #define ZFS_FL_USER_MODIFIABLE (FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL) 980 981 static uint32_t 982 __zpl_ioctl_getflags(struct inode *ip) 983 { 984 uint64_t zfs_flags = ITOZ(ip)->z_pflags; 985 uint32_t ioctl_flags = 0; 986 987 if (zfs_flags & ZFS_IMMUTABLE) 988 ioctl_flags |= FS_IMMUTABLE_FL; 989 990 if (zfs_flags & ZFS_APPENDONLY) 991 ioctl_flags |= FS_APPEND_FL; 992 993 if (zfs_flags & ZFS_NODUMP) 994 ioctl_flags |= FS_NODUMP_FL; 995 996 if (zfs_flags & ZFS_PROJINHERIT) 997 ioctl_flags |= ZFS_PROJINHERIT_FL; 998 999 return (ioctl_flags & ZFS_FL_USER_VISIBLE); 1000 } 1001 1002 /* 1003 * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file 1004 * attributes common to both Linux and Solaris are mapped. 1005 */ 1006 static int 1007 zpl_ioctl_getflags(struct file *filp, void __user *arg) 1008 { 1009 uint32_t flags; 1010 int err; 1011 1012 flags = __zpl_ioctl_getflags(file_inode(filp)); 1013 err = copy_to_user(arg, &flags, sizeof (flags)); 1014 1015 return (err); 1016 } 1017 1018 /* 1019 * fchange() is a helper macro to detect if we have been asked to change a 1020 * flag. This is ugly, but the requirement that we do this is a consequence of 1021 * how the Linux file attribute interface was designed. Another consequence is 1022 * that concurrent modification of files suffers from a TOCTOU race. Neither 1023 * are things we can fix without modifying the kernel-userland interface, which 1024 * is outside of our jurisdiction. 1025 */ 1026 1027 #define fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1))) 1028 1029 static int 1030 __zpl_ioctl_setflags(struct inode *ip, uint32_t ioctl_flags, xvattr_t *xva) 1031 { 1032 uint64_t zfs_flags = ITOZ(ip)->z_pflags; 1033 xoptattr_t *xoap; 1034 1035 if (ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL | 1036 ZFS_PROJINHERIT_FL)) 1037 return (-EOPNOTSUPP); 1038 1039 if (ioctl_flags & ~ZFS_FL_USER_MODIFIABLE) 1040 return (-EACCES); 1041 1042 if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) || 1043 fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) && 1044 !capable(CAP_LINUX_IMMUTABLE)) 1045 return (-EPERM); 1046 1047 if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip)) 1048 return (-EACCES); 1049 1050 xva_init(xva); 1051 xoap = xva_getxoptattr(xva); 1052 1053 #define FLAG_CHANGE(iflag, zflag, xflag, xfield) do { \ 1054 if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) || \ 1055 ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) { \ 1056 XVA_SET_REQ(xva, (xflag)); \ 1057 (xfield) = ((ioctl_flags & (iflag)) != 0); \ 1058 } \ 1059 } while (0) 1060 1061 FLAG_CHANGE(FS_IMMUTABLE_FL, ZFS_IMMUTABLE, XAT_IMMUTABLE, 1062 xoap->xoa_immutable); 1063 FLAG_CHANGE(FS_APPEND_FL, ZFS_APPENDONLY, XAT_APPENDONLY, 1064 xoap->xoa_appendonly); 1065 FLAG_CHANGE(FS_NODUMP_FL, ZFS_NODUMP, XAT_NODUMP, 1066 xoap->xoa_nodump); 1067 FLAG_CHANGE(ZFS_PROJINHERIT_FL, ZFS_PROJINHERIT, XAT_PROJINHERIT, 1068 xoap->xoa_projinherit); 1069 1070 #undef FLAG_CHANGE 1071 1072 return (0); 1073 } 1074 1075 static int 1076 zpl_ioctl_setflags(struct file *filp, void __user *arg) 1077 { 1078 struct inode *ip = file_inode(filp); 1079 uint32_t flags; 1080 cred_t *cr = CRED(); 1081 xvattr_t xva; 1082 int err; 1083 fstrans_cookie_t cookie; 1084 1085 if (copy_from_user(&flags, arg, sizeof (flags))) 1086 return (-EFAULT); 1087 1088 err = __zpl_ioctl_setflags(ip, flags, &xva); 1089 if (err) 1090 return (err); 1091 1092 crhold(cr); 1093 cookie = spl_fstrans_mark(); 1094 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap); 1095 spl_fstrans_unmark(cookie); 1096 crfree(cr); 1097 1098 return (err); 1099 } 1100 1101 static int 1102 zpl_ioctl_getxattr(struct file *filp, void __user *arg) 1103 { 1104 zfsxattr_t fsx = { 0 }; 1105 struct inode *ip = file_inode(filp); 1106 int err; 1107 1108 fsx.fsx_xflags = __zpl_ioctl_getflags(ip); 1109 fsx.fsx_projid = ITOZ(ip)->z_projid; 1110 err = copy_to_user(arg, &fsx, sizeof (fsx)); 1111 1112 return (err); 1113 } 1114 1115 static int 1116 zpl_ioctl_setxattr(struct file *filp, void __user *arg) 1117 { 1118 struct inode *ip = file_inode(filp); 1119 zfsxattr_t fsx; 1120 cred_t *cr = CRED(); 1121 xvattr_t xva; 1122 xoptattr_t *xoap; 1123 int err; 1124 fstrans_cookie_t cookie; 1125 1126 if (copy_from_user(&fsx, arg, sizeof (fsx))) 1127 return (-EFAULT); 1128 1129 if (!zpl_is_valid_projid(fsx.fsx_projid)) 1130 return (-EINVAL); 1131 1132 err = __zpl_ioctl_setflags(ip, fsx.fsx_xflags, &xva); 1133 if (err) 1134 return (err); 1135 1136 xoap = xva_getxoptattr(&xva); 1137 XVA_SET_REQ(&xva, XAT_PROJID); 1138 xoap->xoa_projid = fsx.fsx_projid; 1139 1140 crhold(cr); 1141 cookie = spl_fstrans_mark(); 1142 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap); 1143 spl_fstrans_unmark(cookie); 1144 crfree(cr); 1145 1146 return (err); 1147 } 1148 1149 /* 1150 * Expose Additional File Level Attributes of ZFS. 1151 */ 1152 static int 1153 zpl_ioctl_getdosflags(struct file *filp, void __user *arg) 1154 { 1155 struct inode *ip = file_inode(filp); 1156 uint64_t dosflags = ITOZ(ip)->z_pflags; 1157 dosflags &= ZFS_DOS_FL_USER_VISIBLE; 1158 int err = copy_to_user(arg, &dosflags, sizeof (dosflags)); 1159 1160 return (err); 1161 } 1162 1163 static int 1164 __zpl_ioctl_setdosflags(struct inode *ip, uint64_t ioctl_flags, xvattr_t *xva) 1165 { 1166 uint64_t zfs_flags = ITOZ(ip)->z_pflags; 1167 xoptattr_t *xoap; 1168 1169 if (ioctl_flags & (~ZFS_DOS_FL_USER_VISIBLE)) 1170 return (-EOPNOTSUPP); 1171 1172 if ((fchange(ioctl_flags, zfs_flags, ZFS_IMMUTABLE, ZFS_IMMUTABLE) || 1173 fchange(ioctl_flags, zfs_flags, ZFS_APPENDONLY, ZFS_APPENDONLY)) && 1174 !capable(CAP_LINUX_IMMUTABLE)) 1175 return (-EPERM); 1176 1177 if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip)) 1178 return (-EACCES); 1179 1180 xva_init(xva); 1181 xoap = xva_getxoptattr(xva); 1182 1183 #define FLAG_CHANGE(iflag, xflag, xfield) do { \ 1184 if (((ioctl_flags & (iflag)) && !(zfs_flags & (iflag))) || \ 1185 ((zfs_flags & (iflag)) && !(ioctl_flags & (iflag)))) { \ 1186 XVA_SET_REQ(xva, (xflag)); \ 1187 (xfield) = ((ioctl_flags & (iflag)) != 0); \ 1188 } \ 1189 } while (0) 1190 1191 FLAG_CHANGE(ZFS_IMMUTABLE, XAT_IMMUTABLE, xoap->xoa_immutable); 1192 FLAG_CHANGE(ZFS_APPENDONLY, XAT_APPENDONLY, xoap->xoa_appendonly); 1193 FLAG_CHANGE(ZFS_NODUMP, XAT_NODUMP, xoap->xoa_nodump); 1194 FLAG_CHANGE(ZFS_READONLY, XAT_READONLY, xoap->xoa_readonly); 1195 FLAG_CHANGE(ZFS_HIDDEN, XAT_HIDDEN, xoap->xoa_hidden); 1196 FLAG_CHANGE(ZFS_SYSTEM, XAT_SYSTEM, xoap->xoa_system); 1197 FLAG_CHANGE(ZFS_ARCHIVE, XAT_ARCHIVE, xoap->xoa_archive); 1198 FLAG_CHANGE(ZFS_NOUNLINK, XAT_NOUNLINK, xoap->xoa_nounlink); 1199 FLAG_CHANGE(ZFS_REPARSE, XAT_REPARSE, xoap->xoa_reparse); 1200 FLAG_CHANGE(ZFS_OFFLINE, XAT_OFFLINE, xoap->xoa_offline); 1201 FLAG_CHANGE(ZFS_SPARSE, XAT_SPARSE, xoap->xoa_sparse); 1202 1203 #undef FLAG_CHANGE 1204 1205 return (0); 1206 } 1207 1208 /* 1209 * Set Additional File Level Attributes of ZFS. 1210 */ 1211 static int 1212 zpl_ioctl_setdosflags(struct file *filp, void __user *arg) 1213 { 1214 struct inode *ip = file_inode(filp); 1215 uint64_t dosflags; 1216 cred_t *cr = CRED(); 1217 xvattr_t xva; 1218 int err; 1219 fstrans_cookie_t cookie; 1220 1221 if (copy_from_user(&dosflags, arg, sizeof (dosflags))) 1222 return (-EFAULT); 1223 1224 err = __zpl_ioctl_setdosflags(ip, dosflags, &xva); 1225 if (err) 1226 return (err); 1227 1228 crhold(cr); 1229 cookie = spl_fstrans_mark(); 1230 err = -zfs_setattr(ITOZ(ip), (vattr_t *)&xva, 0, cr, zfs_init_idmap); 1231 spl_fstrans_unmark(cookie); 1232 crfree(cr); 1233 1234 return (err); 1235 } 1236 1237 static long 1238 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 1239 { 1240 switch (cmd) { 1241 case FS_IOC_GETVERSION: 1242 return (zpl_ioctl_getversion(filp, (void *)arg)); 1243 case FS_IOC_GETFLAGS: 1244 return (zpl_ioctl_getflags(filp, (void *)arg)); 1245 case FS_IOC_SETFLAGS: 1246 return (zpl_ioctl_setflags(filp, (void *)arg)); 1247 case ZFS_IOC_FSGETXATTR: 1248 return (zpl_ioctl_getxattr(filp, (void *)arg)); 1249 case ZFS_IOC_FSSETXATTR: 1250 return (zpl_ioctl_setxattr(filp, (void *)arg)); 1251 case ZFS_IOC_GETDOSFLAGS: 1252 return (zpl_ioctl_getdosflags(filp, (void *)arg)); 1253 case ZFS_IOC_SETDOSFLAGS: 1254 return (zpl_ioctl_setdosflags(filp, (void *)arg)); 1255 case ZFS_IOC_COMPAT_FICLONE: 1256 return (zpl_ioctl_ficlone(filp, (void *)arg)); 1257 case ZFS_IOC_COMPAT_FICLONERANGE: 1258 return (zpl_ioctl_ficlonerange(filp, (void *)arg)); 1259 case ZFS_IOC_COMPAT_FIDEDUPERANGE: 1260 return (zpl_ioctl_fideduperange(filp, (void *)arg)); 1261 default: 1262 return (-ENOTTY); 1263 } 1264 } 1265 1266 #ifdef CONFIG_COMPAT 1267 static long 1268 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) 1269 { 1270 switch (cmd) { 1271 case FS_IOC32_GETVERSION: 1272 cmd = FS_IOC_GETVERSION; 1273 break; 1274 case FS_IOC32_GETFLAGS: 1275 cmd = FS_IOC_GETFLAGS; 1276 break; 1277 case FS_IOC32_SETFLAGS: 1278 cmd = FS_IOC_SETFLAGS; 1279 break; 1280 default: 1281 return (-ENOTTY); 1282 } 1283 return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg))); 1284 } 1285 #endif /* CONFIG_COMPAT */ 1286 1287 const struct address_space_operations zpl_address_space_operations = { 1288 #ifdef HAVE_VFS_READPAGES 1289 .readpages = zpl_readpages, 1290 #else 1291 .readahead = zpl_readahead, 1292 #endif 1293 #ifdef HAVE_VFS_READ_FOLIO 1294 .read_folio = zpl_read_folio, 1295 #else 1296 .readpage = zpl_readpage, 1297 #endif 1298 .writepage = zpl_writepage, 1299 .writepages = zpl_writepages, 1300 .direct_IO = zpl_direct_IO, 1301 #ifdef HAVE_VFS_SET_PAGE_DIRTY_NOBUFFERS 1302 .set_page_dirty = __set_page_dirty_nobuffers, 1303 #endif 1304 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO 1305 .dirty_folio = filemap_dirty_folio, 1306 #endif 1307 }; 1308 1309 #ifdef HAVE_VFS_FILE_OPERATIONS_EXTEND 1310 const struct file_operations_extend zpl_file_operations = { 1311 .kabi_fops = { 1312 #else 1313 const struct file_operations zpl_file_operations = { 1314 #endif 1315 .open = zpl_open, 1316 .release = zpl_release, 1317 .llseek = zpl_llseek, 1318 #ifdef HAVE_VFS_RW_ITERATE 1319 #ifdef HAVE_NEW_SYNC_READ 1320 .read = new_sync_read, 1321 .write = new_sync_write, 1322 #endif 1323 .read_iter = zpl_iter_read, 1324 .write_iter = zpl_iter_write, 1325 #ifdef HAVE_VFS_IOV_ITER 1326 #ifdef HAVE_COPY_SPLICE_READ 1327 .splice_read = copy_splice_read, 1328 #else 1329 .splice_read = generic_file_splice_read, 1330 #endif 1331 .splice_write = iter_file_splice_write, 1332 #endif 1333 #else 1334 .read = do_sync_read, 1335 .write = do_sync_write, 1336 .aio_read = zpl_aio_read, 1337 .aio_write = zpl_aio_write, 1338 #endif 1339 .mmap = zpl_mmap, 1340 .fsync = zpl_fsync, 1341 #ifdef HAVE_FILE_AIO_FSYNC 1342 .aio_fsync = zpl_aio_fsync, 1343 #endif 1344 .fallocate = zpl_fallocate, 1345 #ifdef HAVE_VFS_COPY_FILE_RANGE 1346 .copy_file_range = zpl_copy_file_range, 1347 #endif 1348 #ifdef HAVE_VFS_CLONE_FILE_RANGE 1349 .clone_file_range = zpl_clone_file_range, 1350 #endif 1351 #ifdef HAVE_VFS_REMAP_FILE_RANGE 1352 .remap_file_range = zpl_remap_file_range, 1353 #endif 1354 #ifdef HAVE_VFS_DEDUPE_FILE_RANGE 1355 .dedupe_file_range = zpl_dedupe_file_range, 1356 #endif 1357 #ifdef HAVE_FILE_FADVISE 1358 .fadvise = zpl_fadvise, 1359 #endif 1360 .unlocked_ioctl = zpl_ioctl, 1361 #ifdef CONFIG_COMPAT 1362 .compat_ioctl = zpl_compat_ioctl, 1363 #endif 1364 #ifdef HAVE_VFS_FILE_OPERATIONS_EXTEND 1365 }, /* kabi_fops */ 1366 .copy_file_range = zpl_copy_file_range, 1367 .clone_file_range = zpl_clone_file_range, 1368 #endif 1369 }; 1370 1371 const struct file_operations zpl_dir_file_operations = { 1372 .llseek = generic_file_llseek, 1373 .read = generic_read_dir, 1374 #if defined(HAVE_VFS_ITERATE_SHARED) 1375 .iterate_shared = zpl_iterate, 1376 #elif defined(HAVE_VFS_ITERATE) 1377 .iterate = zpl_iterate, 1378 #else 1379 .readdir = zpl_readdir, 1380 #endif 1381 .fsync = zpl_fsync, 1382 .unlocked_ioctl = zpl_ioctl, 1383 #ifdef CONFIG_COMPAT 1384 .compat_ioctl = zpl_compat_ioctl, 1385 #endif 1386 }; 1387 1388 /* CSTYLED */ 1389 module_param(zfs_fallocate_reserve_percent, uint, 0644); 1390 MODULE_PARM_DESC(zfs_fallocate_reserve_percent, 1391 "Percentage of length to use for the available capacity check"); 1392