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.
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.
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]
22 * Copyright (c) 2011, Lawrence Livermore National Security, LLC.
26 #include <sys/dmu_objset.h>
27 #include <sys/zfs_vfsops.h>
28 #include <sys/zfs_vnops.h>
29 #include <sys/zfs_znode.h>
34 zpl_open(struct inode *ip, struct file *filp)
39 error = generic_file_open(ip, filp);
44 error = -zfs_open(ip, filp->f_mode, filp->f_flags, cr);
46 ASSERT3S(error, <=, 0);
52 zpl_release(struct inode *ip, struct file *filp)
57 if (ITOZ(ip)->z_atime_dirty)
61 error = -zfs_close(ip, filp->f_flags, cr);
63 ASSERT3S(error, <=, 0);
69 zpl_iterate(struct file *filp, struct dir_context *ctx)
71 struct dentry *dentry = filp->f_path.dentry;
76 error = -zfs_readdir(dentry->d_inode, ctx, cr);
78 ASSERT3S(error, <=, 0);
83 #if !defined(HAVE_VFS_ITERATE)
85 zpl_readdir(struct file *filp, void *dirent, filldir_t filldir)
87 struct dir_context ctx = DIR_CONTEXT_INIT(dirent, filldir, filp->f_pos);
90 error = zpl_iterate(filp, &ctx);
91 filp->f_pos = ctx.pos;
95 #endif /* HAVE_VFS_ITERATE */
97 #if defined(HAVE_FSYNC_WITH_DENTRY)
99 * Linux 2.6.x - 2.6.34 API,
100 * Through 2.6.34 the nfsd kernel server would pass a NULL 'file struct *'
101 * to the fops->fsync() hook. For this reason, we must be careful not to
102 * use filp unconditionally.
105 zpl_fsync(struct file *filp, struct dentry *dentry, int datasync)
111 error = -zfs_fsync(dentry->d_inode, datasync, cr);
113 ASSERT3S(error, <=, 0);
118 #elif defined(HAVE_FSYNC_WITHOUT_DENTRY)
120 * Linux 2.6.35 - 3.0 API,
121 * As of 2.6.35 the dentry argument to the fops->fsync() hook was deemed
122 * redundant. The dentry is still accessible via filp->f_path.dentry,
123 * and we are guaranteed that filp will never be NULL.
126 zpl_fsync(struct file *filp, int datasync)
128 struct inode *inode = filp->f_mapping->host;
133 error = -zfs_fsync(inode, datasync, cr);
135 ASSERT3S(error, <=, 0);
140 #elif defined(HAVE_FSYNC_RANGE)
142 * Linux 3.1 - 3.x API,
143 * As of 3.1 the responsibility to call filemap_write_and_wait_range() has
144 * been pushed down in to the .fsync() vfs hook. Additionally, the i_mutex
145 * lock is no longer held by the caller, for zfs we don't require the lock
146 * to be held so we don't acquire it.
149 zpl_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
151 struct inode *inode = filp->f_mapping->host;
155 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
160 error = -zfs_fsync(inode, datasync, cr);
162 ASSERT3S(error, <=, 0);
167 #error "Unsupported fops->fsync() implementation"
171 zpl_read_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
172 uio_seg_t segment, int flags, cred_t *cr)
179 iov.iov_base = (void *)buf;
185 uio.uio_loffset = pos;
186 uio.uio_limit = MAXOFFSET_T;
187 uio.uio_segflg = segment;
189 error = -zfs_read(ip, &uio, flags, cr);
193 read = len - uio.uio_resid;
194 task_io_account_read(read);
200 zpl_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
206 read = zpl_read_common(filp->f_mapping->host, buf, len, *ppos,
207 UIO_USERSPACE, filp->f_flags, cr);
218 zpl_write_common(struct inode *ip, const char *buf, size_t len, loff_t pos,
219 uio_seg_t segment, int flags, cred_t *cr)
226 iov.iov_base = (void *)buf;
232 uio.uio_loffset = pos;
233 uio.uio_limit = MAXOFFSET_T;
234 uio.uio_segflg = segment;
236 error = -zfs_write(ip, &uio, flags, cr);
240 wrote = len - uio.uio_resid;
241 task_io_account_write(wrote);
247 zpl_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
253 wrote = zpl_write_common(filp->f_mapping->host, buf, len, *ppos,
254 UIO_USERSPACE, filp->f_flags, cr);
265 zpl_llseek(struct file *filp, loff_t offset, int whence)
267 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
268 if (whence == SEEK_DATA || whence == SEEK_HOLE) {
269 struct inode *ip = filp->f_mapping->host;
270 loff_t maxbytes = ip->i_sb->s_maxbytes;
274 error = -zfs_holey(ip, whence, &offset);
276 error = lseek_execute(filp, ip, offset, maxbytes);
277 spl_inode_unlock(ip);
281 #endif /* SEEK_HOLE && SEEK_DATA */
283 return (generic_file_llseek(filp, offset, whence));
287 * It's worth taking a moment to describe how mmap is implemented
288 * for zfs because it differs considerably from other Linux filesystems.
289 * However, this issue is handled the same way under OpenSolaris.
291 * The issue is that by design zfs bypasses the Linux page cache and
292 * leaves all caching up to the ARC. This has been shown to work
293 * well for the common read(2)/write(2) case. However, mmap(2)
294 * is problem because it relies on being tightly integrated with the
295 * page cache. To handle this we cache mmap'ed files twice, once in
296 * the ARC and a second time in the page cache. The code is careful
297 * to keep both copies synchronized.
299 * When a file with an mmap'ed region is written to using write(2)
300 * both the data in the ARC and existing pages in the page cache
301 * are updated. For a read(2) data will be read first from the page
302 * cache then the ARC if needed. Neither a write(2) or read(2) will
303 * will ever result in new pages being added to the page cache.
305 * New pages are added to the page cache only via .readpage() which
306 * is called when the vfs needs to read a page off disk to back the
307 * virtual memory region. These pages may be modified without
308 * notifying the ARC and will be written out periodically via
309 * .writepage(). This will occur due to either a sync or the usual
310 * page aging behavior. Note because a read(2) of a mmap'ed file
311 * will always check the page cache first even when the ARC is out
312 * of date correct data will still be returned.
314 * While this implementation ensures correct behavior it does have
315 * have some drawbacks. The most obvious of which is that it
316 * increases the required memory footprint when access mmap'ed
317 * files. It also adds additional complexity to the code keeping
318 * both caches synchronized.
320 * Longer term it may be possible to cleanly resolve this wart by
321 * mapping page cache pages directly on to the ARC buffers. The
322 * Linux address space operations are flexible enough to allow
323 * selection of which pages back a particular index. The trick
324 * would be working out the details of which subsystem is in
325 * charge, the ARC, the page cache, or both. It may also prove
326 * helpful to move the ARC buffers to a scatter-gather lists
327 * rather than a vmalloc'ed region.
330 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
332 struct inode *ip = filp->f_mapping->host;
333 znode_t *zp = ITOZ(ip);
336 error = -zfs_map(ip, vma->vm_pgoff, (caddr_t *)vma->vm_start,
337 (size_t)(vma->vm_end - vma->vm_start), vma->vm_flags);
341 error = generic_file_mmap(filp, vma);
345 mutex_enter(&zp->z_lock);
347 mutex_exit(&zp->z_lock);
353 * Populate a page with data for the Linux page cache. This function is
354 * only used to support mmap(2). There will be an identical copy of the
355 * data in the ARC which is kept up to date via .write() and .writepage().
357 * Current this function relies on zpl_read_common() and the O_DIRECT
358 * flag to read in a page. This works but the more correct way is to
359 * update zfs_fillpage() to be Linux friendly and use that interface.
362 zpl_readpage(struct file *filp, struct page *pp)
368 ASSERT(PageLocked(pp));
369 ip = pp->mapping->host;
372 error = -zfs_getpage(ip, pl, 1);
376 ClearPageUptodate(pp);
380 flush_dcache_page(pp);
388 * Populate a set of pages with data for the Linux page cache. This
389 * function will only be called for read ahead and never for demand
390 * paging. For simplicity, the code relies on read_cache_pages() to
391 * correctly lock each page for IO and call zpl_readpage().
394 zpl_readpages(struct file *filp, struct address_space *mapping,
395 struct list_head *pages, unsigned nr_pages)
397 return (read_cache_pages(mapping, pages,
398 (filler_t *)zpl_readpage, filp));
402 zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data)
404 struct address_space *mapping = data;
406 ASSERT(PageLocked(pp));
407 ASSERT(!PageWriteback(pp));
408 ASSERT(!(current->flags & PF_NOFS));
411 * Annotate this call path with a flag that indicates that it is
412 * unsafe to use KM_SLEEP during memory allocations due to the
413 * potential for a deadlock. KM_PUSHPAGE should be used instead.
415 current->flags |= PF_NOFS;
416 (void) zfs_putpage(mapping->host, pp, wbc);
417 current->flags &= ~PF_NOFS;
423 zpl_writepages(struct address_space *mapping, struct writeback_control *wbc)
425 znode_t *zp = ITOZ(mapping->host);
426 zfs_sb_t *zsb = ITOZSB(mapping->host);
427 enum writeback_sync_modes sync_mode;
431 if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
432 wbc->sync_mode = WB_SYNC_ALL;
434 sync_mode = wbc->sync_mode;
437 * We don't want to run write_cache_pages() in SYNC mode here, because
438 * that would make putpage() wait for a single page to be committed to
439 * disk every single time, resulting in atrocious performance. Instead
440 * we run it once in non-SYNC mode so that the ZIL gets all the data,
441 * and then we commit it all in one go.
443 wbc->sync_mode = WB_SYNC_NONE;
444 result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
445 if (sync_mode != wbc->sync_mode) {
448 zil_commit(zsb->z_log, zp->z_id);
452 * We need to call write_cache_pages() again (we can't just
453 * return after the commit) because the previous call in
454 * non-SYNC mode does not guarantee that we got all the dirty
455 * pages (see the implementation of write_cache_pages() for
456 * details). That being said, this is a no-op in most cases.
458 wbc->sync_mode = sync_mode;
459 result = write_cache_pages(mapping, wbc, zpl_putpage, mapping);
465 * Write out dirty pages to the ARC, this function is only required to
466 * support mmap(2). Mapped pages may be dirtied by memory operations
467 * which never call .write(). These dirty pages are kept in sync with
468 * the ARC buffers via this hook.
471 zpl_writepage(struct page *pp, struct writeback_control *wbc)
473 if (ITOZSB(pp->mapping->host)->z_os->os_sync == ZFS_SYNC_ALWAYS)
474 wbc->sync_mode = WB_SYNC_ALL;
476 return (zpl_putpage(pp, wbc, pp->mapping));
480 * The only flag combination which matches the behavior of zfs_space()
481 * is FALLOC_FL_PUNCH_HOLE. This flag was introduced in the 2.6.38 kernel.
484 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
487 int error = -EOPNOTSUPP;
489 if (mode & FALLOC_FL_KEEP_SIZE)
490 return (-EOPNOTSUPP);
494 #ifdef FALLOC_FL_PUNCH_HOLE
495 if (mode & FALLOC_FL_PUNCH_HOLE) {
504 error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
506 #endif /* FALLOC_FL_PUNCH_HOLE */
510 ASSERT3S(error, <=, 0);
514 #ifdef HAVE_FILE_FALLOCATE
516 zpl_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
518 return zpl_fallocate_common(filp->f_path.dentry->d_inode,
521 #endif /* HAVE_FILE_FALLOCATE */
524 zpl_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
527 case ZFS_IOC_GETFLAGS:
528 case ZFS_IOC_SETFLAGS:
529 return (-EOPNOTSUPP);
537 zpl_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
539 return (zpl_ioctl(filp, cmd, arg));
541 #endif /* CONFIG_COMPAT */
544 const struct address_space_operations zpl_address_space_operations = {
545 .readpages = zpl_readpages,
546 .readpage = zpl_readpage,
547 .writepage = zpl_writepage,
548 .writepages = zpl_writepages,
551 const struct file_operations zpl_file_operations = {
553 .release = zpl_release,
554 .llseek = zpl_llseek,
559 #ifdef HAVE_FILE_FALLOCATE
560 .fallocate = zpl_fallocate,
561 #endif /* HAVE_FILE_FALLOCATE */
562 .unlocked_ioctl = zpl_ioctl,
564 .compat_ioctl = zpl_compat_ioctl,
568 const struct file_operations zpl_dir_file_operations = {
569 .llseek = generic_file_llseek,
570 .read = generic_read_dir,
571 #ifdef HAVE_VFS_ITERATE
572 .iterate = zpl_iterate,
574 .readdir = zpl_readdir,
577 .unlocked_ioctl = zpl_ioctl,
579 .compat_ioctl = zpl_compat_ioctl,