blob: 94712fc781fa58fa162c8399c6b9427aa242f5ce [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/file.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
11 *
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
13 *
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
15 *
16 * nfs regular file handling functions
17 */
18
Bryan Schumakerddda8e02012-07-30 16:05:23 -040019#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/time.h>
21#include <linux/kernel.h>
22#include <linux/errno.h>
23#include <linux/fcntl.h>
24#include <linux/stat.h>
25#include <linux/nfs_fs.h>
26#include <linux/nfs_mount.h>
27#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/pagemap.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040029#include <linux/aio.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090030#include <linux/gfp.h>
Trond Myklebustb608b282010-07-30 15:31:54 -040031#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35#include "delegation.h"
Trond Myklebust94387fb2007-07-22 17:09:05 -040036#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050037#include "iostat.h"
David Howells545db452009-04-03 16:42:44 +010038#include "fscache.h"
Christoph Hellwig612aa982014-09-10 08:23:30 -070039#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Trond Myklebustf4ce1292013-08-19 18:59:33 -040041#include "nfstrace.h"
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#define NFSDBG_FACILITY NFSDBG_FILE
44
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040045static const struct vm_operations_struct nfs_file_vm_ops;
Trond Myklebust94387fb2007-07-22 17:09:05 -040046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/* Hack for future NFS swap support */
48#ifndef IS_SWAPFILE
49# define IS_SWAPFILE(inode) (0)
50#endif
51
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -040052int nfs_check_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053{
54 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
55 return -EINVAL;
56
57 return 0;
58}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040059EXPORT_SYMBOL_GPL(nfs_check_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61/*
62 * Open file
63 */
64static int
65nfs_file_open(struct inode *inode, struct file *filp)
66{
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 int res;
68
Al Viro6de14722013-09-16 10:53:17 -040069 dprintk("NFS: open file(%pD2)\n", filp);
Chuck Levercc0dd2d2008-06-11 17:55:42 -040070
Chuck Leverc2459dc2010-02-01 14:17:14 -050071 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 res = nfs_check_flags(filp->f_flags);
73 if (res)
74 return res;
75
Trond Myklebust46cb6502008-06-11 16:32:46 -040076 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 return res;
78}
79
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -040080int
Linus Torvalds1da177e2005-04-16 15:20:36 -070081nfs_file_release(struct inode *inode, struct file *filp)
82{
Al Viro6de14722013-09-16 10:53:17 -040083 dprintk("NFS: release(%pD2)\n", filp);
Chuck Lever6da24bc2008-06-11 17:55:58 -040084
Chuck Lever91d5b472006-03-20 13:44:14 -050085 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Trond Myklebust46cb6502008-06-11 16:32:46 -040086 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040088EXPORT_SYMBOL_GPL(nfs_file_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Trond Myklebust980802e2005-06-13 11:14:01 -040090/**
91 * nfs_revalidate_size - Revalidate the file size
92 * @inode - pointer to inode struct
93 * @file - pointer to struct file
94 *
95 * Revalidates the file length. This is basically a wrapper around
96 * nfs_revalidate_inode() that takes into account the fact that we may
97 * have cached writes (in which case we don't care about the server's
98 * idea of what the file length is), or O_DIRECT (in which case we
99 * shouldn't trust the cache).
100 */
101static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
102{
103 struct nfs_server *server = NFS_SERVER(inode);
104 struct nfs_inode *nfsi = NFS_I(inode);
105
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400106 if (nfs_have_delegated_attributes(inode))
107 goto out_noreval;
108
Trond Myklebust980802e2005-06-13 11:14:01 -0400109 if (filp->f_flags & O_DIRECT)
110 goto force_reval;
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400111 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
112 goto force_reval;
113 if (nfs_attribute_timeout(inode))
114 goto force_reval;
115out_noreval:
116 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400117force_reval:
118 return __nfs_revalidate_inode(server, inode);
119}
120
Andrew Morton965c8e52012-12-17 15:59:39 -0800121loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
Trond Myklebust980802e2005-06-13 11:14:01 -0400122{
Al Viro6de14722013-09-16 10:53:17 -0400123 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
124 filp, offset, whence);
Chuck Leverb84e06c2008-06-11 17:55:34 -0400125
Josef Bacik06222e42011-07-18 13:21:38 -0400126 /*
Andrew Morton965c8e52012-12-17 15:59:39 -0800127 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
Josef Bacik06222e42011-07-18 13:21:38 -0400128 * the cached file length
129 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800130 if (whence != SEEK_SET && whence != SEEK_CUR) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400131 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400132
Trond Myklebust980802e2005-06-13 11:14:01 -0400133 int retval = nfs_revalidate_file_size(inode, filp);
134 if (retval < 0)
135 return (loff_t)retval;
Andi Kleen79835a712011-09-15 16:06:52 -0700136 }
Trond Myklebustd5e66342008-09-23 17:28:35 -0400137
Andrew Morton965c8e52012-12-17 15:59:39 -0800138 return generic_file_llseek(filp, offset, whence);
Trond Myklebust980802e2005-06-13 11:14:01 -0400139}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400140EXPORT_SYMBOL_GPL(nfs_file_llseek);
Trond Myklebust980802e2005-06-13 11:14:01 -0400141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142/*
143 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400145int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700146nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147{
Al Viro6de14722013-09-16 10:53:17 -0400148 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Al Viro6de14722013-09-16 10:53:17 -0400150 dprintk("NFS: flush(%pD2)\n", file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Chuck Leverc2459dc2010-02-01 14:17:14 -0500152 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 if ((file->f_mode & FMODE_WRITE) == 0)
154 return 0;
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400155
Trond Myklebust14546c32012-05-06 19:10:59 -0400156 /*
157 * If we're holding a write delegation, then just start the i/o
158 * but don't wait for completion (or send a commit).
159 */
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400160 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebust14546c32012-05-06 19:10:59 -0400161 return filemap_fdatawrite(file->f_mapping);
162
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400163 /* Flush writes to the server and return any errors */
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400164 return vfs_fsync(file, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400166EXPORT_SYMBOL_GPL(nfs_file_flush);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400168ssize_t
Al Viro3aa2d192014-04-02 20:14:12 -0400169nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Al Viro6de14722013-09-16 10:53:17 -0400171 struct inode *inode = file_inode(iocb->ki_filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 ssize_t result;
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 if (iocb->ki_filp->f_flags & O_DIRECT)
Martin K. Petersene19a8a02014-10-14 09:00:44 -0600175 return nfs_file_direct_read(iocb, to, iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Al Viro619d30b2014-03-04 21:53:33 -0500177 dprintk("NFS: read(%pD2, %zu@%lu)\n",
Al Viro6de14722013-09-16 10:53:17 -0400178 iocb->ki_filp,
Al Viro3aa2d192014-04-02 20:14:12 -0400179 iov_iter_count(to), (unsigned long) iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Trond Myklebust44b11872006-05-25 01:40:59 -0400181 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500182 if (!result) {
Al Viro3aa2d192014-04-02 20:14:12 -0400183 result = generic_file_read_iter(iocb, to);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500184 if (result > 0)
185 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 return result;
188}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400189EXPORT_SYMBOL_GPL(nfs_file_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400191ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200192nfs_file_splice_read(struct file *filp, loff_t *ppos,
193 struct pipe_inode_info *pipe, size_t count,
194 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
Al Viro6de14722013-09-16 10:53:17 -0400196 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 ssize_t res;
198
Al Viro6de14722013-09-16 10:53:17 -0400199 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
200 filp, (unsigned long) count, (unsigned long long) *ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Trond Myklebust44b11872006-05-25 01:40:59 -0400202 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500203 if (!res) {
Jens Axboef0930ff2007-06-01 11:51:43 +0200204 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500205 if (res > 0)
206 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 return res;
209}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400210EXPORT_SYMBOL_GPL(nfs_file_splice_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400212int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
214{
Al Viro6de14722013-09-16 10:53:17 -0400215 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 int status;
217
Al Viro6de14722013-09-16 10:53:17 -0400218 dprintk("NFS: mmap(%pD2)\n", file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400220 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
221 * so we call that before revalidating the mapping
222 */
223 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400224 if (!status) {
225 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400226 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return status;
229}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400230EXPORT_SYMBOL_GPL(nfs_file_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232/*
233 * Flush any dirty pages for this process, and check for write errors.
234 * The return status from this call provides a reliable indication of
235 * whether any write errors occurred for this process.
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400236 *
237 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
238 * disk, but it retrieves and clears ctx->error after synching, despite
239 * the two being set at the same time in nfs_context_set_write_error().
240 * This is because the former is used to notify the _next_ call to
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300241 * nfs_file_write() that a write error occurred, and hence cause it to
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400242 * fall back to doing a synchronous write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400244int
Bryan Schumakera5c58892012-06-20 15:53:42 -0400245nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400247 struct nfs_open_context *ctx = nfs_file_open_context(file);
Al Viro6de14722013-09-16 10:53:17 -0400248 struct inode *inode = file_inode(file);
Trond Myklebust05990d12012-09-11 16:01:22 -0400249 int have_error, do_resend, status;
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400250 int ret = 0;
251
Al Viro6de14722013-09-16 10:53:17 -0400252 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Chuck Lever91d5b472006-03-20 13:44:14 -0500254 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebust05990d12012-09-11 16:01:22 -0400255 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400256 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
257 status = nfs_commit_inode(inode, FLUSH_SYNC);
258 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
Trond Myklebust05990d12012-09-11 16:01:22 -0400259 if (have_error) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400260 ret = xchg(&ctx->error, 0);
Trond Myklebust05990d12012-09-11 16:01:22 -0400261 if (ret)
262 goto out;
263 }
264 if (status < 0) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400265 ret = status;
Trond Myklebust05990d12012-09-11 16:01:22 -0400266 goto out;
267 }
268 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
269 if (do_resend)
270 ret = -EAGAIN;
271out:
Bryan Schumakera5c58892012-06-20 15:53:42 -0400272 return ret;
273}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400274EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
Bryan Schumakera5c58892012-06-20 15:53:42 -0400275
276static int
277nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
278{
279 int ret;
Al Viro496ad9a2013-01-23 17:07:38 -0500280 struct inode *inode = file_inode(file);
Bryan Schumakera5c58892012-06-20 15:53:42 -0400281
Trond Myklebustf4ce1292013-08-19 18:59:33 -0400282 trace_nfs_fsync_enter(inode);
283
Trond Myklebust05990d12012-09-11 16:01:22 -0400284 do {
285 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
286 if (ret != 0)
287 break;
288 mutex_lock(&inode->i_mutex);
289 ret = nfs_file_fsync_commit(file, start, end, datasync);
290 mutex_unlock(&inode->i_mutex);
Trond Myklebustdcfc4f22012-09-11 16:19:38 -0400291 /*
292 * If nfs_file_fsync_commit detected a server reboot, then
293 * resend all dirty pages that might have been covered by
294 * the NFS_CONTEXT_RESEND_WRITES flag
295 */
296 start = 0;
297 end = LLONG_MAX;
Trond Myklebust05990d12012-09-11 16:01:22 -0400298 } while (ret == -EAGAIN);
299
Trond Myklebustf4ce1292013-08-19 18:59:33 -0400300 trace_nfs_fsync_exit(inode, ret);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400301 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303
304/*
Peter Staubach38c73042009-08-10 08:54:16 -0400305 * Decide whether a read/modify/write cycle may be more efficient
306 * then a modify/write/read cycle when writing to a page in the
307 * page cache.
308 *
309 * The modify/write/read cycle may occur if a page is read before
310 * being completely filled by the writer. In this situation, the
311 * page must be completely written to stable storage on the server
312 * before it can be refilled by reading in the page from the server.
313 * This can lead to expensive, small, FILE_SYNC mode writes being
314 * done.
315 *
316 * It may be more efficient to read the page first if the file is
317 * open for reading in addition to writing, the page is not marked
318 * as Uptodate, it is not dirty or waiting to be committed,
319 * indicating that it was previously allocated and then modified,
320 * that there were valid bytes of data in that range of the file,
321 * and that the new data won't completely replace the old data in
322 * that range of the file.
323 */
324static int nfs_want_read_modify_write(struct file *file, struct page *page,
325 loff_t pos, unsigned len)
326{
327 unsigned int pglen = nfs_page_length(page);
328 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
329 unsigned int end = offset + len;
330
Christoph Hellwig612aa982014-09-10 08:23:30 -0700331 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
332 if (!PageUptodate(page))
333 return 1;
334 return 0;
335 }
336
Peter Staubach38c73042009-08-10 08:54:16 -0400337 if ((file->f_mode & FMODE_READ) && /* open for read? */
338 !PageUptodate(page) && /* Uptodate? */
339 !PagePrivate(page) && /* i/o request already? */
340 pglen && /* valid bytes of file? */
341 (end < pglen || offset)) /* replace all valid bytes? */
342 return 1;
343 return 0;
344}
345
346/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700347 * This does the "real" work of the write. We must allocate and lock the
348 * page to be sent back to the generic routine, which then copies the
349 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 *
351 * If the writer ends up delaying the write, the writer needs to
352 * increment the page use counts until he is done with the page.
353 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700354static int nfs_write_begin(struct file *file, struct address_space *mapping,
355 loff_t pos, unsigned len, unsigned flags,
356 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700358 int ret;
Peter Staubach38c73042009-08-10 08:54:16 -0400359 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700360 struct page *page;
Peter Staubach38c73042009-08-10 08:54:16 -0400361 int once_thru = 0;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700362
Niels de Vos1e8968c2013-12-17 18:20:16 +0100363 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400364 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400365
Peter Staubach38c73042009-08-10 08:54:16 -0400366start:
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400367 /*
368 * Prevent starvation issues if someone is doing a consistency
369 * sync-to-disk
370 */
NeilBrown74316202014-07-07 15:16:04 +1000371 ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
372 nfs_wait_bit_killable, TASK_KILLABLE);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400373 if (ret)
374 return ret;
375
Nick Piggin54566b22009-01-04 12:00:53 -0800376 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700377 if (!page)
378 return -ENOMEM;
379 *pagep = page;
380
381 ret = nfs_flush_incompatible(file, page);
382 if (ret) {
383 unlock_page(page);
384 page_cache_release(page);
Peter Staubach38c73042009-08-10 08:54:16 -0400385 } else if (!once_thru &&
386 nfs_want_read_modify_write(file, page, pos, len)) {
387 once_thru = 1;
388 ret = nfs_readpage(file, page);
389 page_cache_release(page);
390 if (!ret)
391 goto start;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700392 }
393 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
Nick Piggin4899f9c2007-10-16 01:25:16 -0700396static int nfs_write_end(struct file *file, struct address_space *mapping,
397 loff_t pos, unsigned len, unsigned copied,
398 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700400 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
Andy Adamsondc248262013-08-14 11:59:16 -0400401 struct nfs_open_context *ctx = nfs_file_open_context(file);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700402 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Niels de Vos1e8968c2013-12-17 18:20:16 +0100404 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400405 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400406
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400407 /*
408 * Zero any uninitialised parts of the page, and then mark the page
409 * as up to date if it turns out that we're extending the file.
410 */
411 if (!PageUptodate(page)) {
412 unsigned pglen = nfs_page_length(page);
413 unsigned end = offset + len;
414
415 if (pglen == 0) {
416 zero_user_segments(page, 0, offset,
417 end, PAGE_CACHE_SIZE);
418 SetPageUptodate(page);
419 } else if (end >= pglen) {
420 zero_user_segment(page, end, PAGE_CACHE_SIZE);
421 if (offset == 0)
422 SetPageUptodate(page);
423 } else
424 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
425 }
426
Nick Piggin4899f9c2007-10-16 01:25:16 -0700427 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700428
429 unlock_page(page);
430 page_cache_release(page);
431
Chuck Lever3d509e52007-12-20 14:55:04 -0500432 if (status < 0)
433 return status;
Andy Adamson2701d082012-05-24 13:13:24 -0400434 NFS_I(mapping->host)->write_io += copied;
Andy Adamsondc248262013-08-14 11:59:16 -0400435
436 if (nfs_ctx_key_to_expire(ctx)) {
437 status = nfs_wb_all(mapping->host);
438 if (status < 0)
439 return status;
440 }
441
Chuck Lever3d509e52007-12-20 14:55:04 -0500442 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
David Howells6b9b3512009-04-03 16:42:41 +0100445/*
446 * Partially or wholly invalidate a page
447 * - Release the private state associated with a page if undergoing complete
448 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100449 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100450 * - Caller holds page lock
451 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400452static void nfs_invalidate_page(struct page *page, unsigned int offset,
453 unsigned int length)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500454{
Lukas Czernerd47992f2013-05-21 23:17:23 -0400455 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
456 page, offset, length);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400457
Lukas Czernerd47992f2013-05-21 23:17:23 -0400458 if (offset != 0 || length < PAGE_CACHE_SIZE)
Trond Myklebust1c759502006-10-09 16:18:38 -0400459 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400460 /* Cancel any unstarted writes on this page */
Mel Gormand56b4dd2012-07-31 16:45:06 -0700461 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
David Howells545db452009-04-03 16:42:44 +0100462
463 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500464}
465
David Howells6b9b3512009-04-03 16:42:41 +0100466/*
467 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100468 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100469 * - Caller holds page lock
470 * - Return true (may release page) or false (may not)
471 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500472static int nfs_release_page(struct page *page, gfp_t gfp)
473{
Trond Myklebustb608b282010-07-30 15:31:54 -0400474 struct address_space *mapping = page->mapping;
475
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400476 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
477
NeilBrown95905442014-09-24 11:28:32 +1000478 /* Always try to initiate a 'commit' if relevant, but only
NeilBrown1aff5252014-09-24 11:28:32 +1000479 * wait for it if __GFP_WAIT is set. Even then, only wait 1
480 * second and only if the 'bdi' is not congested.
NeilBrown95905442014-09-24 11:28:32 +1000481 * Waiting indefinitely can cause deadlocks when the NFS
NeilBrown1aff5252014-09-24 11:28:32 +1000482 * server is on this machine, when a new TCP connection is
483 * needed and in other rare cases. There is no particular
484 * need to wait extensively here. A short wait has the
485 * benefit that someone else can worry about the freezer.
Jeff Layton5cf02d02012-07-23 13:58:51 -0400486 */
NeilBrown95905442014-09-24 11:28:32 +1000487 if (mapping) {
488 struct nfs_server *nfss = NFS_SERVER(mapping->host);
489 nfs_commit_inode(mapping->host, 0);
490 if ((gfp & __GFP_WAIT) &&
NeilBrown353db792014-09-24 11:28:32 +1000491 !bdi_write_congested(&nfss->backing_dev_info)) {
NeilBrown95905442014-09-24 11:28:32 +1000492 wait_on_page_bit_killable_timeout(page, PG_private,
493 HZ);
NeilBrown353db792014-09-24 11:28:32 +1000494 if (PagePrivate(page))
495 set_bdi_congested(&nfss->backing_dev_info,
496 BLK_RW_ASYNC);
NeilBrown95905442014-09-24 11:28:32 +1000497 }
Trond Myklebustb608b282010-07-30 15:31:54 -0400498 }
Trond Myklebuste3db7692007-01-10 23:15:39 -0800499 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100500 if (PagePrivate(page))
501 return 0;
502 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800503}
504
Mel Gormanf919b192013-07-03 15:02:06 -0700505static void nfs_check_dirty_writeback(struct page *page,
506 bool *dirty, bool *writeback)
507{
508 struct nfs_inode *nfsi;
509 struct address_space *mapping = page_file_mapping(page);
510
511 if (!mapping || PageSwapCache(page))
512 return;
513
514 /*
515 * Check if an unstable page is currently being committed and
516 * if so, have the VM treat it as if the page is under writeback
517 * so it will not block due to pages that will shortly be freeable.
518 */
519 nfsi = NFS_I(mapping->host);
520 if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
521 *writeback = true;
522 return;
523 }
524
525 /*
526 * If PagePrivate() is set, then the page is not freeable and as the
527 * inode is not being committed, it's not going to be cleaned in the
528 * near future so treat it as dirty
529 */
530 if (PagePrivate(page))
531 *dirty = true;
532}
533
David Howells6b9b3512009-04-03 16:42:41 +0100534/*
535 * Attempt to clear the private state associated with a page when an error
536 * occurs that requires the cached contents of an inode to be written back or
537 * destroyed
David Howells545db452009-04-03 16:42:44 +0100538 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100539 * - Caller holds page lock
540 * - Return 0 if successful, -error otherwise
541 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800542static int nfs_launder_page(struct page *page)
543{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700544 struct inode *inode = page_file_mapping(page)->host;
David Howells545db452009-04-03 16:42:44 +0100545 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400546
547 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
548 inode->i_ino, (long long)page_offset(page));
549
David Howells545db452009-04-03 16:42:44 +0100550 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400551 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500552}
553
Mel Gormana564b8f2012-07-31 16:45:12 -0700554#ifdef CONFIG_NFS_SWAP
555static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
556 sector_t *span)
557{
Jeff Laytondad2b012014-09-10 09:03:55 -0400558 int ret;
559 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
560
Mel Gormana564b8f2012-07-31 16:45:12 -0700561 *span = sis->pages;
Jeff Laytondad2b012014-09-10 09:03:55 -0400562
563 rcu_read_lock();
564 ret = xs_swapper(rcu_dereference(clnt->cl_xprt), 1);
565 rcu_read_unlock();
566
567 return ret;
Mel Gormana564b8f2012-07-31 16:45:12 -0700568}
569
570static void nfs_swap_deactivate(struct file *file)
571{
Jeff Laytondad2b012014-09-10 09:03:55 -0400572 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
573
574 rcu_read_lock();
575 xs_swapper(rcu_dereference(clnt->cl_xprt), 0);
576 rcu_read_unlock();
Mel Gormana564b8f2012-07-31 16:45:12 -0700577}
578#endif
579
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700580const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 .readpage = nfs_readpage,
582 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400583 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 .writepage = nfs_writepage,
585 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700586 .write_begin = nfs_write_begin,
587 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500588 .invalidatepage = nfs_invalidate_page,
589 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 .direct_IO = nfs_direct_IO,
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400591 .migratepage = nfs_migrate_page,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800592 .launder_page = nfs_launder_page,
Mel Gormanf919b192013-07-03 15:02:06 -0700593 .is_dirty_writeback = nfs_check_dirty_writeback,
Andi Kleenf590f332009-09-16 11:50:17 +0200594 .error_remove_page = generic_error_remove_page,
Mel Gormana564b8f2012-07-31 16:45:12 -0700595#ifdef CONFIG_NFS_SWAP
596 .swap_activate = nfs_swap_activate,
597 .swap_deactivate = nfs_swap_deactivate,
598#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599};
600
David Howells6b9b3512009-04-03 16:42:41 +0100601/*
602 * Notification that a PTE pointing to an NFS page is about to be made
603 * writable, implying that someone is about to modify the page through a
604 * shared-writable mapping
605 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700606static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400607{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700608 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400609 struct file *filp = vma->vm_file;
Al Viro6de14722013-09-16 10:53:17 -0400610 struct inode *inode = file_inode(filp);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400611 unsigned pagelen;
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400612 int ret = VM_FAULT_NOPAGE;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700613 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400614
Niels de Vos1e8968c2013-12-17 18:20:16 +0100615 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400616 filp, filp->f_mapping->host->i_ino,
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400617 (long long)page_offset(page));
618
David Howells545db452009-04-03 16:42:44 +0100619 /* make sure the cache has finished storing the page */
Al Viro6de14722013-09-16 10:53:17 -0400620 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
David Howells545db452009-04-03 16:42:44 +0100621
Trond Myklebust94387fb2007-07-22 17:09:05 -0400622 lock_page(page);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700623 mapping = page_file_mapping(page);
Al Viro6de14722013-09-16 10:53:17 -0400624 if (mapping != inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500625 goto out_unlock;
626
Trond Myklebust2aeb98f2012-01-17 22:04:26 -0500627 wait_on_page_writeback(page);
628
Trond Myklebust94387fb2007-07-22 17:09:05 -0400629 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500630 if (pagelen == 0)
631 goto out_unlock;
632
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400633 ret = VM_FAULT_LOCKED;
634 if (nfs_flush_incompatible(filp, page) == 0 &&
635 nfs_updatepage(filp, page, 0, pagelen) == 0)
636 goto out;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500637
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400638 ret = VM_FAULT_SIGBUS;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500639out_unlock:
Trond Myklebust94387fb2007-07-22 17:09:05 -0400640 unlock_page(page);
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400641out:
642 return ret;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400643}
644
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400645static const struct vm_operations_struct nfs_file_vm_ops = {
Trond Myklebust94387fb2007-07-22 17:09:05 -0400646 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700647 .map_pages = filemap_map_pages,
Trond Myklebust94387fb2007-07-22 17:09:05 -0400648 .page_mkwrite = nfs_vm_page_mkwrite,
649};
650
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400651static int nfs_need_sync_write(struct file *filp, struct inode *inode)
652{
653 struct nfs_open_context *ctx;
654
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100655 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400656 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400657 ctx = nfs_file_open_context(filp);
Andy Adamsondc248262013-08-14 11:59:16 -0400658 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
659 nfs_ctx_key_to_expire(ctx))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400660 return 1;
661 return 0;
662}
663
Al Viroedaf4362014-04-03 14:07:25 -0400664ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Al Viro6de14722013-09-16 10:53:17 -0400666 struct file *file = iocb->ki_filp;
667 struct inode *inode = file_inode(file);
Chuck Lever7e381172010-02-01 14:17:41 -0500668 unsigned long written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 ssize_t result;
Al Viroedaf4362014-04-03 14:07:25 -0400670 size_t count = iov_iter_count(from);
671 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Al Viro6de14722013-09-16 10:53:17 -0400673 result = nfs_key_timeout_notify(file, inode);
Andy Adamsondc248262013-08-14 11:59:16 -0400674 if (result)
675 return result;
676
Al Viro6de14722013-09-16 10:53:17 -0400677 if (file->f_flags & O_DIRECT)
Martin K. Petersene19a8a02014-10-14 09:00:44 -0600678 return nfs_file_direct_write(iocb, from, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Al Viro619d30b2014-03-04 21:53:33 -0500680 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
681 file, count, (long long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
683 result = -EBUSY;
684 if (IS_SWAPFILE(inode))
685 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000686 /*
687 * O_APPEND implies that we must revalidate the file length.
688 */
Al Viro6de14722013-09-16 10:53:17 -0400689 if (file->f_flags & O_APPEND) {
690 result = nfs_revalidate_file_size(inode, file);
Trond Myklebust7d52e862005-06-22 17:16:30 +0000691 if (result)
692 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
695 result = count;
696 if (!count)
697 goto out;
698
Al Viroedaf4362014-04-03 14:07:25 -0400699 result = generic_file_write_iter(iocb, from);
Chuck Lever7e381172010-02-01 14:17:41 -0500700 if (result > 0)
701 written = result;
702
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100703 /* Return error values for O_DSYNC and IS_SYNC() */
Al Viro6de14722013-09-16 10:53:17 -0400704 if (result >= 0 && nfs_need_sync_write(file, inode)) {
705 int err = vfs_fsync(file, 0);
Trond Myklebust200baa22006-12-05 00:35:40 -0500706 if (err < 0)
707 result = err;
708 }
Chuck Lever7e381172010-02-01 14:17:41 -0500709 if (result > 0)
710 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711out:
712 return result;
713
714out_swapfile:
715 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
716 goto out;
717}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400718EXPORT_SYMBOL_GPL(nfs_file_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400720static int
721do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722{
723 struct inode *inode = filp->f_mapping->host;
724 int status = 0;
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000725 unsigned int saved_type = fl->fl_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Trond Myklebust039c4d72005-10-18 14:20:16 -0700727 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400728 posix_test_lock(filp, fl);
729 if (fl->fl_type != F_UNLCK) {
730 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700731 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 }
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000733 fl->fl_type = saved_type;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700734
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400735 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
Trond Myklebust039c4d72005-10-18 14:20:16 -0700736 goto out_noconflict;
737
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400738 if (is_local)
Trond Myklebust039c4d72005-10-18 14:20:16 -0700739 goto out_noconflict;
740
741 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
742out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700744out_noconflict:
745 fl->fl_type = F_UNLCK;
746 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
749static int do_vfs_lock(struct file *file, struct file_lock *fl)
750{
751 int res = 0;
752 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
753 case FL_POSIX:
754 res = posix_lock_file_wait(file, fl);
755 break;
756 case FL_FLOCK:
757 res = flock_lock_file_wait(file, fl);
758 break;
759 default:
760 BUG();
761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return res;
763}
764
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400765static int
766do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
768 struct inode *inode = filp->f_mapping->host;
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400769 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 int status;
771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 /*
773 * Flush all pending writes before doing anything
774 * with locks..
775 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500776 nfs_sync_mapping(filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400778 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
779 if (!IS_ERR(l_ctx)) {
780 status = nfs_iocounter_wait(&l_ctx->io_count);
781 nfs_put_lock_context(l_ctx);
782 if (status < 0)
783 return status;
784 }
785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 /* NOTE: special case
787 * If we're signalled while cleaning up locks on process exit, we
788 * still need to complete the unlock.
789 */
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400790 /*
791 * Use local locking if mounted with "-onolock" or with appropriate
792 * "-olocal_lock="
793 */
794 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
796 else
797 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 return status;
799}
800
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400801static int
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700802is_time_granular(struct timespec *ts) {
803 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
804}
805
806static int
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400807do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
809 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 int status;
811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 /*
813 * Flush all pending writes before doing anything
814 * with locks..
815 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500816 status = nfs_sync_mapping(filp->f_mapping);
817 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 goto out;
819
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400820 /*
821 * Use local locking if mounted with "-onolock" or with appropriate
822 * "-olocal_lock="
823 */
824 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400826 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 if (status < 0)
829 goto out;
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 /*
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700832 * Revalidate the cache if the server has time stamps granular
833 * enough to detect subsecond changes. Otherwise, clear the
834 * cache to prevent missing any changes.
835 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 * This makes locking act as a cache coherency point.
837 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500838 nfs_sync_mapping(filp->f_mapping);
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400839 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700840 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
841 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
842 else
843 nfs_zap_caches(inode);
844 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 return status;
847}
848
849/*
850 * Lock a (portion of) a file
851 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400852int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400854 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400855 int ret = -ENOLCK;
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400856 int is_local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Al Viro6de14722013-09-16 10:53:17 -0400858 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
859 filp, fl->fl_type, fl->fl_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400861
Chuck Lever91d5b472006-03-20 13:44:14 -0500862 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700865 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400866 goto out_err;
867
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400868 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
869 is_local = 1;
870
Trond Myklebust21162712008-05-20 19:34:39 -0400871 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
872 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
873 if (ret < 0)
874 goto out_err;
875 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
877 if (IS_GETLK(cmd))
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400878 ret = do_getlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400879 else if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400880 ret = do_unlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400881 else
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400882 ret = do_setlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400883out_err:
884 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400886EXPORT_SYMBOL_GPL(nfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888/*
889 * Lock a (portion of) a file
890 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400891int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400893 struct inode *inode = filp->f_mapping->host;
894 int is_local = 0;
895
Al Viro6de14722013-09-16 10:53:17 -0400896 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
897 filp, fl->fl_type, fl->fl_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (!(fl->fl_flags & FL_FLOCK))
900 return -ENOLCK;
901
Jeff Laytonad0fcd42012-07-23 15:46:23 -0400902 /*
903 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
904 * any standard. In principle we might be able to support LOCK_MAND
905 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
906 * NFS code is not set up for it.
907 */
908 if (fl->fl_type & LOCK_MAND)
909 return -EINVAL;
910
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400911 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
912 is_local = 1;
913
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 /* We're simulating flock() locks using posix locks on the server */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400916 return do_unlk(filp, cmd, fl, is_local);
917 return do_setlk(filp, cmd, fl, is_local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400919EXPORT_SYMBOL_GPL(nfs_flock);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400920
Jeff Layton04869582011-11-04 13:31:22 -0400921const struct file_operations nfs_file_operations = {
922 .llseek = nfs_file_llseek,
Al Viro3aa2d192014-04-02 20:14:12 -0400923 .read = new_sync_read,
Al Viroedaf4362014-04-03 14:07:25 -0400924 .write = new_sync_write,
Al Viro3aa2d192014-04-02 20:14:12 -0400925 .read_iter = nfs_file_read,
Al Viroedaf4362014-04-03 14:07:25 -0400926 .write_iter = nfs_file_write,
Jeff Layton04869582011-11-04 13:31:22 -0400927 .mmap = nfs_file_mmap,
928 .open = nfs_file_open,
929 .flush = nfs_file_flush,
930 .release = nfs_file_release,
931 .fsync = nfs_file_fsync,
932 .lock = nfs_lock,
933 .flock = nfs_flock,
934 .splice_read = nfs_file_splice_read,
Al Viro4da54c22014-04-05 04:37:17 -0400935 .splice_write = iter_file_splice_write,
Jeff Layton04869582011-11-04 13:31:22 -0400936 .check_flags = nfs_check_flags,
Jeff Layton1c994a02014-08-27 06:49:41 -0400937 .setlease = simple_nosetlease,
Jeff Layton04869582011-11-04 13:31:22 -0400938};
Bryan Schumakerddda8e02012-07-30 16:05:23 -0400939EXPORT_SYMBOL_GPL(nfs_file_operations);