module/zfs/zpl_file.c

   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 http://www.opensolaris.org/os/licensing.
  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 <sys/file.h>
  31 #include <sys/dmu_objset.h>
  32 #include <sys/zfs_vfsops.h>
  33 #include <sys/zfs_vnops.h>
  34 #include <sys/zfs_znode.h>
  35 #include <sys/zfs_project.h>
  36
  37
  38 static int
  39 zpl_open(struct inode *ip, struct file *filp)
  40 {
  41         cred_t *cr = CRED();
  42         int error;
  43         fstrans_cookie_t cookie;
  44
  45         error = generic_file_open(ip, filp);
  46         if (error)
  47                 return (error);
  48
  49         crhold(cr);
  50         cookie = spl_fstrans_mark();
  51         error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
  52         spl_fstrans_unmark(cookie);
  53         crfree(cr);
  54         ASSERT3S(error, <=, 0);
  55
  56         return (error);
  57 }
  58
  59 static int
  60 zpl_release(struct inode *ip, struct file *filp)
  61 {
  62         cred_t *cr = CRED();
  63         int error;
  64         fstrans_cookie_t cookie;
  65
  66         cookie = spl_fstrans_mark();
  67         if (ITOZ(ip)->z_atime_dirty)
  68                 zfs_mark_inode_dirty(ip);
  69
  70         crhold(cr);
  71         error = -zfs_close(ip, filp->f_flags, cr);
  72         spl_fstrans_unmark(cookie);
  73         crfree(cr);
  74         ASSERT3S(error, <=, 0);
  75
  76         return (error);
  77 }
  78
  79 static int
  80 zpl_iterate(struct file *filp, zpl_dir_context_t *ctx)
  81 {
  82         cred_t *cr = CRED();
  83         int error;
  84         fstrans_cookie_t cookie;
  85
  86         crhold(cr);
  87         cookie = spl_fstrans_mark();
  88         error = -zfs_readdir(file_inode(filp), ctx, cr);
  89         spl_fstrans_unmark(cookie);
  90         crfree(cr);
  91         ASSERT3S(error, <=, 0);
  92
  93         return (error);
  94 }
  95
  96 #if !defined(HAVE_VFS_ITERATE) && !defined(HAVE_VFS_ITERATE_SHARED)
  97 static int
  98 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir)
  99 {
 100         zpl_dir_context_t ctx =
 101             ZPL_DIR_CONTEXT_INIT(dirent, filldir, filp->f_pos);
 102         int error;
 103
 104         error = zpl_iterate(filp, &ctx);
 105         filp->f_pos = ctx.pos;
 106
 107         return (error);
 108 }
 109 #endif /* !HAVE_VFS_ITERATE && !HAVE_VFS_ITERATE_SHARED */
 110
 111 #if defined(HAVE_FSYNC_WITH_DENTRY)
 112 /*
 113  * Linux 2.6.x - 2.6.34 API,
 114  * Through 2.6.34 the nfsd kernel server would pass a NULL 'file struct *'
 115  * to the fops->fsync() hook.  For this reason, we must be careful not to
 116  * use filp unconditionally.
 117  */
 118 static int
 119 zpl_fsync(struct file *filp, struct dentry *dentry, int datasync)
 120 {
 121         cred_t *cr = CRED();
 122         int error;
 123         fstrans_cookie_t cookie;
 124
 125         crhold(cr);
 126         cookie = spl_fstrans_mark();
 127         error = -zfs_fsync(dentry->d_inode, datasync, cr);
 128         spl_fstrans_unmark(cookie);
 129         crfree(cr);
 130         ASSERT3S(error, <=, 0);
 131
 132         return (error);
 133 }
 134
 135 #ifdef HAVE_FILE_AIO_FSYNC
 136 static int
 137 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 138 {
 139         struct file *filp = kiocb->ki_filp;
 140         return (zpl_fsync(filp, file_dentry(filp), datasync));
 141 }
 142 #endif
 143
 144 #elif defined(HAVE_FSYNC_WITHOUT_DENTRY)
 145 /*
 146  * Linux 2.6.35 - 3.0 API,
 147  * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
 148  * redundant.  The dentry is still accessible via filp->f_path.dentry,
 149  * and we are guaranteed that filp will never be NULL.
 150  */
 151 static int
 152 zpl_fsync(struct file *filp, int datasync)
 153 {
 154         struct inode *inode = filp->f_mapping->host;
 155         cred_t *cr = CRED();
 156         int error;
 157         fstrans_cookie_t cookie;
 158
 159         crhold(cr);
 160         cookie = spl_fstrans_mark();
 161         error = -zfs_fsync(inode, datasync, cr);
 162         spl_fstrans_unmark(cookie);
 163         crfree(cr);
 164         ASSERT3S(error, <=, 0);
 165
 166         return (error);
 167 }
 168
 169 #ifdef HAVE_FILE_AIO_FSYNC
 170 static int
 171 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 172 {
 173         return (zpl_fsync(kiocb->ki_filp, datasync));
 174 }
 175 #endif
 176
 177 #elif defined(HAVE_FSYNC_RANGE)
 178 /*
 179  * Linux 3.1 - 3.x API,
 180  * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
 181  * been pushed down in to the .fsync() vfs hook.  Additionally, the i_mutex
 182  * lock is no longer held by the caller, for zfs we don't require the lock
 183  * to be held so we don't acquire it.
 184  */
 185 static int
 186 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 187 {
 188         struct inode *inode = filp->f_mapping->host;
 189         cred_t *cr = CRED();
 190         int error;
 191         fstrans_cookie_t cookie;
 192
 193         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
 194         if (error)
 195                 return (error);
 196
 197         crhold(cr);
 198         cookie = spl_fstrans_mark();
 199         error = -zfs_fsync(inode, datasync, cr);
 200         spl_fstrans_unmark(cookie);
 201         crfree(cr);
 202         ASSERT3S(error, <=, 0);
 203
 204         return (error);
 205 }
 206
 207 #ifdef HAVE_FILE_AIO_FSYNC
 208 static int
 209 zpl_aio_fsync(struct kiocb *kiocb, int datasync)
 210 {
 211         return (zpl_fsync(kiocb->ki_filp, kiocb->ki_pos, -1, datasync));
 212 }
 213 #endif
 214
 215 #else
 216 #error "Unsupported fops->fsync() implementation"
 217 #endif
 218
 219 static ssize_t
 220 zpl_read_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 221     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 222     cred_t *cr, size_t skip)
 223 {
 224         ssize_t read;
 225         uio_t uio;
 226         int error;
 227         fstrans_cookie_t cookie;
 228
 229         uio.uio_iov = iovp;
 230         uio.uio_skip = skip;
 231         uio.uio_resid = count;
 232         uio.uio_iovcnt = nr_segs;
 233         uio.uio_loffset = *ppos;
 234         uio.uio_limit = MAXOFFSET_T;
 235         uio.uio_segflg = segment;
 236
 237         cookie = spl_fstrans_mark();
 238         error = -zfs_read(ip, &uio, flags, cr);
 239         spl_fstrans_unmark(cookie);
 240         if (error < 0)
 241                 return (error);
 242
 243         read = count - uio.uio_resid;
 244         *ppos += read;
 245         task_io_account_read(read);
 246
 247         return (read);
 248 }
 249
 250 inline ssize_t
 251 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 252     uio_seg_t segment, int flags, cred_t *cr)
 253 {
 254         struct iovec iov;
 255
 256         iov.iov_base = (void *)buf;
 257         iov.iov_len = len;
 258
 259         return (zpl_read_common_iovec(ip, &iov, len, 1, ppos, segment,
 260             flags, cr, 0));
 261 }
 262
 263 static ssize_t
 264 zpl_iter_read_common(struct kiocb *kiocb, const struct iovec *iovp,
 265     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 266 {
 267         cred_t *cr = CRED();
 268         struct file *filp = kiocb->ki_filp;
 269         ssize_t read;
 270
 271         crhold(cr);
 272         read = zpl_read_common_iovec(filp->f_mapping->host, iovp, count,
 273             nr_segs, &kiocb->ki_pos, seg, filp->f_flags, cr, skip);
 274         crfree(cr);
 275
 276         file_accessed(filp);
 277         return (read);
 278 }
 279
 280 #if defined(HAVE_VFS_RW_ITERATE)
 281 static ssize_t
 282 zpl_iter_read(struct kiocb *kiocb, struct iov_iter *to)
 283 {
 284         ssize_t ret;
 285         uio_seg_t seg = UIO_USERSPACE;
 286         if (to->type & ITER_KVEC)
 287                 seg = UIO_SYSSPACE;
 288         if (to->type & ITER_BVEC)
 289                 seg = UIO_BVEC;
 290         ret = zpl_iter_read_common(kiocb, to->iov, to->nr_segs,
 291             iov_iter_count(to), seg, to->iov_offset);
 292         if (ret > 0)
 293                 iov_iter_advance(to, ret);
 294         return (ret);
 295 }
 296 #else
 297 static ssize_t
 298 zpl_aio_read(struct kiocb *kiocb, const struct iovec *iovp,
 299     unsigned long nr_segs, loff_t pos)
 300 {
 301         ssize_t ret;
 302         size_t count;
 303
 304         ret = generic_segment_checks(iovp, &nr_segs, &count, VERIFY_WRITE);
 305         if (ret)
 306                 return (ret);
 307
 308         return (zpl_iter_read_common(kiocb, iovp, nr_segs, count,
 309             UIO_USERSPACE, 0));
 310 }
 311 #endif /* HAVE_VFS_RW_ITERATE */
 312
 313 static ssize_t
 314 zpl_write_common_iovec(struct inode *ip, const struct iovec *iovp, size_t count,
 315     unsigned long nr_segs, loff_t *ppos, uio_seg_t segment, int flags,
 316     cred_t *cr, size_t skip)
 317 {
 318         ssize_t wrote;
 319         uio_t uio;
 320         int error;
 321         fstrans_cookie_t cookie;
 322
 323         if (flags & O_APPEND)
 324                 *ppos = i_size_read(ip);
 325
 326         uio.uio_iov = iovp;
 327         uio.uio_skip = skip;
 328         uio.uio_resid = count;
 329         uio.uio_iovcnt = nr_segs;
 330         uio.uio_loffset = *ppos;
 331         uio.uio_limit = MAXOFFSET_T;
 332         uio.uio_segflg = segment;
 333
 334         cookie = spl_fstrans_mark();
 335         error = -zfs_write(ip, &uio, flags, cr);
 336         spl_fstrans_unmark(cookie);
 337         if (error < 0)
 338                 return (error);
 339
 340         wrote = count - uio.uio_resid;
 341         *ppos += wrote;
 342         task_io_account_write(wrote);
 343
 344         return (wrote);
 345 }
 346
 347 inline ssize_t
 348 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t *ppos,
 349     uio_seg_t segment, int flags, cred_t *cr)
 350 {
 351         struct iovec iov;
 352
 353         iov.iov_base = (void *)buf;
 354         iov.iov_len = len;
 355
 356         return (zpl_write_common_iovec(ip, &iov, len, 1, ppos, segment,
 357             flags, cr, 0));
 358 }
 359
 360 static ssize_t
 361 zpl_iter_write_common(struct kiocb *kiocb, const struct iovec *iovp,
 362     unsigned long nr_segs, size_t count, uio_seg_t seg, size_t skip)
 363 {
 364         cred_t *cr = CRED();
 365         struct file *filp = kiocb->ki_filp;
 366         ssize_t wrote;
 367
 368         crhold(cr);
 369         wrote = zpl_write_common_iovec(filp->f_mapping->host, iovp, count,
 370             nr_segs, &kiocb->ki_pos, seg, filp->f_flags, cr, skip);
 371         crfree(cr);
 372
 373         return (wrote);
 374 }
 375
 376 #if defined(HAVE_VFS_RW_ITERATE)
 377 static ssize_t
 378 zpl_iter_write(struct kiocb *kiocb, struct iov_iter *from)
 379 {
 380         size_t count;
 381         ssize_t ret;
 382         uio_seg_t seg = UIO_USERSPACE;
 383
 384 #ifndef HAVE_GENERIC_WRITE_CHECKS_KIOCB
 385         struct file *file = kiocb->ki_filp;
 386         struct address_space *mapping = file->f_mapping;
 387         struct inode *ip = mapping->host;
 388         int isblk = S_ISBLK(ip->i_mode);
 389
 390         count = iov_iter_count(from);
 391         ret = generic_write_checks(file, &kiocb->ki_pos, &count, isblk);
 392         if (ret)
 393                 return (ret);
 394 #else
 395         /*
 396          * XXX - ideally this check should be in the same lock region with
 397          * write operations, so that there's no TOCTTOU race when doing
 398          * append and someone else grow the file.
 399          */
 400         ret = generic_write_checks(kiocb, from);
 401         if (ret <= 0)
 402                 return (ret);
 403         count = ret;
 404 #endif
 405
 406         if (from->type & ITER_KVEC)
 407                 seg = UIO_SYSSPACE;
 408         if (from->type & ITER_BVEC)
 409                 seg = UIO_BVEC;
 410
 411         ret = zpl_iter_write_common(kiocb, from->iov, from->nr_segs,
 412             count, seg, from->iov_offset);
 413         if (ret > 0)
 414                 iov_iter_advance(from, ret);
 415
 416         return (ret);
 417 }
 418 #else
 419 static ssize_t
 420 zpl_aio_write(struct kiocb *kiocb, const struct iovec *iovp,
 421     unsigned long nr_segs, loff_t pos)
 422 {
 423         struct file *file = kiocb->ki_filp;
 424         struct address_space *mapping = file->f_mapping;
 425         struct inode *ip = mapping->host;
 426         int isblk = S_ISBLK(ip->i_mode);
 427         size_t count;
 428         ssize_t ret;
 429
 430         ret = generic_segment_checks(iovp, &nr_segs, &count, VERIFY_READ);
 431         if (ret)
 432                 return (ret);
 433
 434         ret = generic_write_checks(file, &pos, &count, isblk);
 435         if (ret)
 436                 return (ret);
 437
 438         return (zpl_iter_write_common(kiocb, iovp, nr_segs, count,
 439             UIO_USERSPACE, 0));
 440 }
 441 #endif /* HAVE_VFS_RW_ITERATE */
 442
 443 static loff_t
 444 zpl_llseek(struct file *filp, loff_t offset, int whence)
 445 {
 446 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
 447         fstrans_cookie_t cookie;
 448
 449         if (whence == SEEK_DATA || whence == SEEK_HOLE) {
 450                 struct inode *ip = filp->f_mapping->host;
 451                 loff_t maxbytes = ip->i_sb->s_maxbytes;
 452                 loff_t error;
 453
 454                 spl_inode_lock_shared(ip);
 455                 cookie = spl_fstrans_mark();
 456                 error = -zfs_holey(ip, whence, &offset);
 457                 spl_fstrans_unmark(cookie);
 458                 if (error == 0)
 459                         error = lseek_execute(filp, ip, offset, maxbytes);
 460                 spl_inode_unlock_shared(ip);
 461
 462                 return (error);
 463         }
 464 #endif /* SEEK_HOLE && SEEK_DATA */
 465
 466         return (generic_file_llseek(filp, offset, whence));
 467 }
 468
 469 /*
 470  * It's worth taking a moment to describe how mmap is implemented
 471  * for zfs because it differs considerably from other Linux filesystems.
 472  * However, this issue is handled the same way under OpenSolaris.
 473  *
 474  * The issue is that by design zfs bypasses the Linux page cache and
 475  * leaves all caching up to the ARC.  This has been shown to work
 476  * well for the common read(2)/write(2) case.  However, mmap(2)
 477  * is problem because it relies on being tightly integrated with the
 478  * page cache.  To handle this we cache mmap'ed files twice, once in
 479  * the ARC and a second time in the page cache.  The code is careful
 480  * to keep both copies synchronized.
 481  *
 482  * When a file with an mmap'ed region is written to using write(2)
 483  * both the data in the ARC and existing pages in the page cache
 484  * are updated.  For a read(2) data will be read first from the page
 485  * cache then the ARC if needed.  Neither a write(2) or read(2) will
 486  * will ever result in new pages being added to the page cache.
 487  *
 488  * New pages are added to the page cache only via .readpage() which
 489  * is called when the vfs needs to read a page off disk to back the
 490  * virtual memory region.  These pages may be modified without
 491  * notifying the ARC and will be written out periodically via
 492  * .writepage().  This will occur due to either a sync or the usual
 493  * page aging behavior.  Note because a read(2) of a mmap'ed file
 494  * will always check the page cache first even when the ARC is out
 495  * of date correct data will still be returned.
 496  *
 497  * While this implementation ensures correct behavior it does have
 498  * have some drawbacks.  The most obvious of which is that it
 499  * increases the required memory footprint when access mmap'ed
 500  * files.  It also adds additional complexity to the code keeping
 501  * both caches synchronized.
 502  *
 503  * Longer term it may be possible to cleanly resolve this wart by
 504  * mapping page cache pages directly on to the ARC buffers.  The
 505  * Linux address space operations are flexible enough to allow
 506  * selection of which pages back a particular index.  The trick
 507  * would be working out the details of which subsystem is in
 508  * charge, the ARC, the page cache, or both.  It may also prove
 509  * helpful to move the ARC buffers to a scatter-gather lists
 510  * rather than a vmalloc'ed region.
 511  */
 512 static int
 513 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 514 {
 515         struct inode *ip = filp->f_mapping->host;
 516         znode_t *zp = ITOZ(ip);
 517         int error;
 518         fstrans_cookie_t cookie;
 519
 520         cookie = spl_fstrans_mark();
 521         error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
 522             (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
 523         spl_fstrans_unmark(cookie);
 524         if (error)
 525                 return (error);
 526
 527         error = generic_file_mmap(filp, vma);
 528         if (error)
 529                 return (error);
 530
 531         mutex_enter(&zp->z_lock);
 532         zp->z_is_mapped = B_TRUE;
 533         mutex_exit(&zp->z_lock);
 534
 535         return (error);
 536 }
 537
 538 /*
 539  * Populate a page with data for the Linux page cache.  This function is
 540  * only used to support mmap(2).  There will be an identical copy of the
 541  * data in the ARC which is kept up to date via .write() and .writepage().
 542  *
 543  * Current this function relies on zpl_read_common() and the O_DIRECT
 544  * flag to read in a page.  This works but the more correct way is to
 545  * update zfs_fillpage() to be Linux friendly and use that interface.
 546  */
 547 static int
 548 zpl_readpage(struct file *filp, struct page *pp)
 549 {
 550         struct inode *ip;
 551         struct page *pl[1];
 552         int error = 0;
 553         fstrans_cookie_t cookie;
 554
 555         ASSERT(PageLocked(pp));
 556         ip = pp->mapping->host;
 557         pl[0] = pp;
 558
 559         cookie = spl_fstrans_mark();
 560         error = -zfs_getpage(ip, pl, 1);
 561         spl_fstrans_unmark(cookie);
 562
 563         if (error) {
 564                 SetPageError(pp);
 565                 ClearPageUptodate(pp);
 566         } else {
 567                 ClearPageError(pp);
 568                 SetPageUptodate(pp);
 569                 flush_dcache_page(pp);
 570         }
 571
 572         unlock_page(pp);
 573         return (error);
 574 }
 575
 576 /*
 577  * Populate a set of pages with data for the Linux page cache.  This
 578  * function will only be called for read ahead and never for demand
 579  * paging.  For simplicity, the code relies on read_cache_pages() to
 580  * correctly lock each page for IO and call zpl_readpage().
 581  */
 582 static int
 583 zpl_readpages(struct file *filp, struct address_space *mapping,
 584     struct list_head *pages, unsigned nr_pages)
 585 {
 586         return (read_cache_pages(mapping, pages,
 587             (filler_t *)zpl_readpage, filp));
 588 }
 589
 590 int
 591 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
 592 {
 593         struct address_space *mapping = data;
 594         fstrans_cookie_t cookie;
 595
 596         ASSERT(PageLocked(pp));
 597         ASSERT(!PageWriteback(pp));
 598
 599         cookie = spl_fstrans_mark();
 600         (void) zfs_putpage(mapping->host, pp, wbc);
 601         spl_fstrans_unmark(cookie);
 602
 603         return (0);
 604 }
 605
 606 static int
 607 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
 608 {
 609         znode_t         *zp = ITOZ(mapping->host);
 610         zfsvfs_t        *zfsvfs = ITOZSB(mapping->host);
 611         enum writeback_sync_modes sync_mode;
 612         int result;
 613
 614         ZFS_ENTER(zfsvfs);
 615         if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 616                 wbc->sync_mode = WB_SYNC_ALL;
 617         ZFS_EXIT(zfsvfs);
 618         sync_mode = wbc->sync_mode;
 619
 620         /*
 621          * We don't want to run write_cache_pages() in SYNC mode here, because
 622          * that would make putpage() wait for a single page to be committed to
 623          * disk every single time, resulting in atrocious performance. Instead
 624          * we run it once in non-SYNC mode so that the ZIL gets all the data,
 625          * and then we commit it all in one go.
 626          */
 627         wbc->sync_mode = WB_SYNC_NONE;
 628         result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 629         if (sync_mode != wbc->sync_mode) {
 630                 ZFS_ENTER(zfsvfs);
 631                 ZFS_VERIFY_ZP(zp);
 632                 if (zfsvfs->z_log != NULL)
 633                         zil_commit(zfsvfs->z_log, zp->z_id);
 634                 ZFS_EXIT(zfsvfs);
 635
 636                 /*
 637                  * We need to call write_cache_pages() again (we can't just
 638                  * return after the commit) because the previous call in
 639                  * non-SYNC mode does not guarantee that we got all the dirty
 640                  * pages (see the implementation of write_cache_pages() for
 641                  * details). That being said, this is a no-op in most cases.
 642                  */
 643                 wbc->sync_mode = sync_mode;
 644                 result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
 645         }
 646         return (result);
 647 }
 648
 649 /*
 650  * Write out dirty pages to the ARC, this function is only required to
 651  * support mmap(2).  Mapped pages may be dirtied by memory operations
 652  * which never call .write().  These dirty pages are kept in sync with
 653  * the ARC buffers via this hook.
 654  */
 655 static int
 656 zpl_writepage(struct page *pp, struct writeback_control *wbc)
 657 {
 658         if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
 659                 wbc->sync_mode = WB_SYNC_ALL;
 660
 661         return (zpl_putpage(pp, wbc, pp->mapping));
 662 }
 663
 664 /*
 665  * The only flag combination which matches the behavior of zfs_space()
 666  * is FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
 667  * flag was introduced in the 2.6.38 kernel.
 668  */
 669 #if defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE)
 670 long
 671 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
 672 {
 673         int error = -EOPNOTSUPP;
 674
 675 #if defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
 676         cred_t *cr = CRED();
 677         flock64_t bf;
 678         loff_t olen;
 679         fstrans_cookie_t cookie;
 680
 681         if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 682                 return (error);
 683
 684         if (offset < 0 || len <= 0)
 685                 return (-EINVAL);
 686
 687         spl_inode_lock(ip);
 688         olen = i_size_read(ip);
 689
 690         if (offset > olen) {
 691                 spl_inode_unlock(ip);
 692                 return (0);
 693         }
 694         if (offset + len > olen)
 695                 len = olen - offset;
 696         bf.l_type = F_WRLCK;
 697         bf.l_whence = 0;
 698         bf.l_start = offset;
 699         bf.l_len = len;
 700         bf.l_pid = 0;
 701
 702         crhold(cr);
 703         cookie = spl_fstrans_mark();
 704         error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
 705         spl_fstrans_unmark(cookie);
 706         spl_inode_unlock(ip);
 707
 708         crfree(cr);
 709 #endif /* defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE) */
 710
 711         ASSERT3S(error, <=, 0);
 712         return (error);
 713 }
 714 #endif /* defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE) */
 715
 716 #ifdef HAVE_FILE_FALLOCATE
 717 static long
 718 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
 719 {
 720         return zpl_fallocate_common(file_inode(filp),
 721             mode, offset, len);
 722 }
 723 #endif /* HAVE_FILE_FALLOCATE */
 724
 725 #define ZFS_FL_USER_VISIBLE     (FS_FL_USER_VISIBLE | ZFS_PROJINHERIT_FL)
 726 #define ZFS_FL_USER_MODIFIABLE  (FS_FL_USER_MODIFIABLE | ZFS_PROJINHERIT_FL)
 727
 728 static uint32_t
 729 __zpl_ioctl_getflags(struct inode *ip)
 730 {
 731         uint64_t zfs_flags = ITOZ(ip)->z_pflags;
 732         uint32_t ioctl_flags = 0;
 733
 734         if (zfs_flags & ZFS_IMMUTABLE)
 735                 ioctl_flags |= FS_IMMUTABLE_FL;
 736
 737         if (zfs_flags & ZFS_APPENDONLY)
 738                 ioctl_flags |= FS_APPEND_FL;
 739
 740         if (zfs_flags & ZFS_NODUMP)
 741                 ioctl_flags |= FS_NODUMP_FL;
 742
 743         if (zfs_flags & ZFS_PROJINHERIT)
 744                 ioctl_flags |= ZFS_PROJINHERIT_FL;
 745
 746         return (ioctl_flags & ZFS_FL_USER_VISIBLE);
 747 }
 748
 749 /*
 750  * Map zfs file z_pflags (xvattr_t) to linux file attributes. Only file
 751  * attributes common to both Linux and Solaris are mapped.
 752  */
 753 static int
 754 zpl_ioctl_getflags(struct file *filp, void __user *arg)
 755 {
 756         uint32_t flags;
 757         int err;
 758
 759         flags = __zpl_ioctl_getflags(file_inode(filp));
 760         err = copy_to_user(arg, &flags, sizeof (flags));
 761
 762         return (err);
 763 }
 764
 765 /*
 766  * fchange() is a helper macro to detect if we have been asked to change a
 767  * flag. This is ugly, but the requirement that we do this is a consequence of
 768  * how the Linux file attribute interface was designed. Another consequence is
 769  * that concurrent modification of files suffers from a TOCTOU race. Neither
 770  * are things we can fix without modifying the kernel-userland interface, which
 771  * is outside of our jurisdiction.
 772  */
 773
 774 #define fchange(f0, f1, b0, b1) (!((f0) & (b0)) != !((f1) & (b1)))
 775
 776 static int
 777 __zpl_ioctl_setflags(struct inode *ip, uint32_t ioctl_flags, xvattr_t *xva)
 778 {
 779         uint64_t zfs_flags = ITOZ(ip)->z_pflags;
 780         xoptattr_t *xoap;
 781
 782         if (ioctl_flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NODUMP_FL |
 783             ZFS_PROJINHERIT_FL))
 784                 return (-EOPNOTSUPP);
 785
 786         if (ioctl_flags & ~ZFS_FL_USER_MODIFIABLE)
 787                 return (-EACCES);
 788
 789         if ((fchange(ioctl_flags, zfs_flags, FS_IMMUTABLE_FL, ZFS_IMMUTABLE) ||
 790             fchange(ioctl_flags, zfs_flags, FS_APPEND_FL, ZFS_APPENDONLY)) &&
 791             !capable(CAP_LINUX_IMMUTABLE))
 792                 return (-EACCES);
 793
 794         if (!zpl_inode_owner_or_capable(ip))
 795                 return (-EACCES);
 796
 797         xva_init(xva);
 798         xoap = xva_getxoptattr(xva);
 799
 800         XVA_SET_REQ(xva, XAT_IMMUTABLE);
 801         if (ioctl_flags & FS_IMMUTABLE_FL)
 802                 xoap->xoa_immutable = B_TRUE;
 803
 804         XVA_SET_REQ(xva, XAT_APPENDONLY);
 805         if (ioctl_flags & FS_APPEND_FL)
 806                 xoap->xoa_appendonly = B_TRUE;
 807
 808         XVA_SET_REQ(xva, XAT_NODUMP);
 809         if (ioctl_flags & FS_NODUMP_FL)
 810                 xoap->xoa_nodump = B_TRUE;
 811
 812         XVA_SET_REQ(xva, XAT_PROJINHERIT);
 813         if (ioctl_flags & ZFS_PROJINHERIT_FL)
 814                 xoap->xoa_projinherit = B_TRUE;
 815
 816         return (0);
 817 }
 818
 819 static int
 820 zpl_ioctl_setflags(struct file *filp, void __user *arg)
 821 {
 822         struct inode *ip = file_inode(filp);
 823         uint32_t flags;
 824         cred_t *cr = CRED();
 825         xvattr_t xva;
 826         int err;
 827         fstrans_cookie_t cookie;
 828
 829         if (copy_from_user(&flags, arg, sizeof (flags)))
 830                 return (-EFAULT);
 831
 832         err = __zpl_ioctl_setflags(ip, flags, &xva);
 833         if (err)
 834                 return (err);
 835
 836         crhold(cr);
 837         cookie = spl_fstrans_mark();
 838         err = -zfs_setattr(ip, (vattr_t *)&xva, 0, cr);
 839         spl_fstrans_unmark(cookie);
 840         crfree(cr);
 841
 842         return (err);
 843 }
 844
 845 static int
 846 zpl_ioctl_getxattr(struct file *filp, void __user *arg)
 847 {
 848         zfsxattr_t fsx = { 0 };
 849         struct inode *ip = file_inode(filp);
 850         int err;
 851
 852         fsx.fsx_xflags = __zpl_ioctl_getflags(ip);
 853         fsx.fsx_projid = ITOZ(ip)->z_projid;
 854         err = copy_to_user(arg, &fsx, sizeof (fsx));
 855
 856         return (err);
 857 }
 858
 859 static int
 860 zpl_ioctl_setxattr(struct file *filp, void __user *arg)
 861 {
 862         struct inode *ip = file_inode(filp);
 863         zfsxattr_t fsx;
 864         cred_t *cr = CRED();
 865         xvattr_t xva;
 866         xoptattr_t *xoap;
 867         int err;
 868         fstrans_cookie_t cookie;
 869
 870         if (copy_from_user(&fsx, arg, sizeof (fsx)))
 871                 return (-EFAULT);
 872
 873         if (!zpl_is_valid_projid(fsx.fsx_projid))
 874                 return (-EINVAL);
 875
 876         err = __zpl_ioctl_setflags(ip, fsx.fsx_xflags, &xva);
 877         if (err)
 878                 return (err);
 879
 880         xoap = xva_getxoptattr(&xva);
 881         XVA_SET_REQ(&xva, XAT_PROJID);
 882         xoap->xoa_projid = fsx.fsx_projid;
 883
 884         crhold(cr);
 885         cookie = spl_fstrans_mark();
 886         err = -zfs_setattr(ip, (vattr_t *)&xva, 0, cr);
 887         spl_fstrans_unmark(cookie);
 888         crfree(cr);
 889
 890         return (err);
 891 }
 892
 893 static long
 894 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 895 {
 896         switch (cmd) {
 897         case FS_IOC_GETFLAGS:
 898                 return (zpl_ioctl_getflags(filp, (void *)arg));
 899         case FS_IOC_SETFLAGS:
 900                 return (zpl_ioctl_setflags(filp, (void *)arg));
 901         case ZFS_IOC_FSGETXATTR:
 902                 return (zpl_ioctl_getxattr(filp, (void *)arg));
 903         case ZFS_IOC_FSSETXATTR:
 904                 return (zpl_ioctl_setxattr(filp, (void *)arg));
 905         default:
 906                 return (-ENOTTY);
 907         }
 908 }
 909
 910 #ifdef CONFIG_COMPAT
 911 static long
 912 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 913 {
 914         switch (cmd) {
 915         case FS_IOC32_GETFLAGS:
 916                 cmd = FS_IOC_GETFLAGS;
 917                 break;
 918         case FS_IOC32_SETFLAGS:
 919                 cmd = FS_IOC_SETFLAGS;
 920                 break;
 921         default:
 922                 return (-ENOTTY);
 923         }
 924         return (zpl_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)));
 925 }
 926 #endif /* CONFIG_COMPAT */
 927
 928
 929 const struct address_space_operations zpl_address_space_operations = {
 930         .readpages      = zpl_readpages,
 931         .readpage       = zpl_readpage,
 932         .writepage      = zpl_writepage,
 933         .writepages     = zpl_writepages,
 934 };
 935
 936 const struct file_operations zpl_file_operations = {
 937         .open           = zpl_open,
 938         .release        = zpl_release,
 939         .llseek         = zpl_llseek,
 940 #ifdef HAVE_VFS_RW_ITERATE
 941 #ifdef HAVE_NEW_SYNC_READ
 942         .read           = new_sync_read,
 943         .write          = new_sync_write,
 944 #endif
 945         .read_iter      = zpl_iter_read,
 946         .write_iter     = zpl_iter_write,
 947 #else
 948         .read           = do_sync_read,
 949         .write          = do_sync_write,
 950         .aio_read       = zpl_aio_read,
 951         .aio_write      = zpl_aio_write,
 952 #endif
 953         .mmap           = zpl_mmap,
 954         .fsync          = zpl_fsync,
 955 #ifdef HAVE_FILE_AIO_FSYNC
 956         .aio_fsync      = zpl_aio_fsync,
 957 #endif
 958 #ifdef HAVE_FILE_FALLOCATE
 959         .fallocate      = zpl_fallocate,
 960 #endif /* HAVE_FILE_FALLOCATE */
 961         .unlocked_ioctl = zpl_ioctl,
 962 #ifdef CONFIG_COMPAT
 963         .compat_ioctl   = zpl_compat_ioctl,
 964 #endif
 965 };
 966
 967 const struct file_operations zpl_dir_file_operations = {
 968         .llseek         = generic_file_llseek,
 969         .read           = generic_read_dir,
 970 #if defined(HAVE_VFS_ITERATE_SHARED)
 971         .iterate_shared = zpl_iterate,
 972 #elif defined(HAVE_VFS_ITERATE)
 973         .iterate        = zpl_iterate,
 974 #else
 975         .readdir        = zpl_readdir,
 976 #endif
 977         .fsync          = zpl_fsync,
 978         .unlocked_ioctl = zpl_ioctl,
 979 #ifdef CONFIG_COMPAT
 980         .compat_ioctl   = zpl_compat_ioctl,
 981 #endif
 982 };