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