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