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