blob: 284ca901fe16cd26d064c603f2e51441708a5953 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Trond Myklebustf4ce1292013-08-19 18:59:33 -040040#include "nfstrace.h"
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#define NFSDBG_FACILITY NFSDBG_FILE
43
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040044static const struct vm_operations_struct nfs_file_vm_ops;
Trond Myklebust94387fb2007-07-22 17:09:05 -040045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046/* Hack for future NFS swap support */
47#ifndef IS_SWAPFILE
48# define IS_SWAPFILE(inode) (0)
49#endif
50
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -040051int nfs_check_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
54 return -EINVAL;
55
56 return 0;
57}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040058EXPORT_SYMBOL_GPL(nfs_check_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
61 * Open file
62 */
63static int
64nfs_file_open(struct inode *inode, struct file *filp)
65{
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 int res;
67
Al Viro6de14722013-09-16 10:53:17 -040068 dprintk("NFS: open file(%pD2)\n", filp);
Chuck Levercc0dd2d2008-06-11 17:55:42 -040069
Chuck Leverc2459dc2010-02-01 14:17:14 -050070 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 res = nfs_check_flags(filp->f_flags);
72 if (res)
73 return res;
74
Trond Myklebust46cb6502008-06-11 16:32:46 -040075 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 return res;
77}
78
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -040079int
Linus Torvalds1da177e2005-04-16 15:20:36 -070080nfs_file_release(struct inode *inode, struct file *filp)
81{
Al Viro6de14722013-09-16 10:53:17 -040082 dprintk("NFS: release(%pD2)\n", filp);
Chuck Lever6da24bc2008-06-11 17:55:58 -040083
Chuck Lever91d5b472006-03-20 13:44:14 -050084 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Trond Myklebust46cb6502008-06-11 16:32:46 -040085 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040087EXPORT_SYMBOL_GPL(nfs_file_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Trond Myklebust980802e2005-06-13 11:14:01 -040089/**
90 * nfs_revalidate_size - Revalidate the file size
91 * @inode - pointer to inode struct
92 * @file - pointer to struct file
93 *
94 * Revalidates the file length. This is basically a wrapper around
95 * nfs_revalidate_inode() that takes into account the fact that we may
96 * have cached writes (in which case we don't care about the server's
97 * idea of what the file length is), or O_DIRECT (in which case we
98 * shouldn't trust the cache).
99 */
100static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101{
102 struct nfs_server *server = NFS_SERVER(inode);
103 struct nfs_inode *nfsi = NFS_I(inode);
104
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400105 if (nfs_have_delegated_attributes(inode))
106 goto out_noreval;
107
Trond Myklebust980802e2005-06-13 11:14:01 -0400108 if (filp->f_flags & O_DIRECT)
109 goto force_reval;
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400110 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
111 goto force_reval;
112 if (nfs_attribute_timeout(inode))
113 goto force_reval;
114out_noreval:
115 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400116force_reval:
117 return __nfs_revalidate_inode(server, inode);
118}
119
Andrew Morton965c8e52012-12-17 15:59:39 -0800120loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
Trond Myklebust980802e2005-06-13 11:14:01 -0400121{
Al Viro6de14722013-09-16 10:53:17 -0400122 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
123 filp, offset, whence);
Chuck Leverb84e06c2008-06-11 17:55:34 -0400124
Josef Bacik06222e42011-07-18 13:21:38 -0400125 /*
Andrew Morton965c8e52012-12-17 15:59:39 -0800126 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
Josef Bacik06222e42011-07-18 13:21:38 -0400127 * the cached file length
128 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800129 if (whence != SEEK_SET && whence != SEEK_CUR) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400130 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400131
Trond Myklebust980802e2005-06-13 11:14:01 -0400132 int retval = nfs_revalidate_file_size(inode, filp);
133 if (retval < 0)
134 return (loff_t)retval;
Andi Kleen79835a712011-09-15 16:06:52 -0700135 }
Trond Myklebustd5e66342008-09-23 17:28:35 -0400136
Andrew Morton965c8e52012-12-17 15:59:39 -0800137 return generic_file_llseek(filp, offset, whence);
Trond Myklebust980802e2005-06-13 11:14:01 -0400138}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400139EXPORT_SYMBOL_GPL(nfs_file_llseek);
Trond Myklebust980802e2005-06-13 11:14:01 -0400140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141/*
142 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400144int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700145nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
Al Viro6de14722013-09-16 10:53:17 -0400147 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Al Viro6de14722013-09-16 10:53:17 -0400149 dprintk("NFS: flush(%pD2)\n", file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Chuck Leverc2459dc2010-02-01 14:17:14 -0500151 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 if ((file->f_mode & FMODE_WRITE) == 0)
153 return 0;
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400154
Trond Myklebust14546c32012-05-06 19:10:59 -0400155 /*
156 * If we're holding a write delegation, then just start the i/o
157 * but don't wait for completion (or send a commit).
158 */
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400159 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebust14546c32012-05-06 19:10:59 -0400160 return filemap_fdatawrite(file->f_mapping);
161
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400162 /* Flush writes to the server and return any errors */
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400163 return vfs_fsync(file, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400165EXPORT_SYMBOL_GPL(nfs_file_flush);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400167ssize_t
Badari Pulavarty027445c2006-09-30 23:28:46 -0700168nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
169 unsigned long nr_segs, loff_t pos)
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)
Mel Gormana564b8f2012-07-31 16:45:12 -0700175 return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Al Viro6de14722013-09-16 10:53:17 -0400177 dprintk("NFS: read(%pD2, %lu@%lu)\n",
178 iocb->ki_filp,
Rakib Mullick6f276e42011-11-01 12:16:15 +0600179 (unsigned long) iov_length(iov, nr_segs), (unsigned long) 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) {
Badari Pulavarty027445c2006-09-30 23:28:46 -0700183 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
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
331 if ((file->f_mode & FMODE_READ) && /* open for read? */
332 !PageUptodate(page) && /* Uptodate? */
333 !PagePrivate(page) && /* i/o request already? */
334 pglen && /* valid bytes of file? */
335 (end < pglen || offset)) /* replace all valid bytes? */
336 return 1;
337 return 0;
338}
339
340/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700341 * This does the "real" work of the write. We must allocate and lock the
342 * page to be sent back to the generic routine, which then copies the
343 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 *
345 * If the writer ends up delaying the write, the writer needs to
346 * increment the page use counts until he is done with the page.
347 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700348static int nfs_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700352 int ret;
Peter Staubach38c73042009-08-10 08:54:16 -0400353 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700354 struct page *page;
Peter Staubach38c73042009-08-10 08:54:16 -0400355 int once_thru = 0;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700356
Niels de Vos1e8968c2013-12-17 18:20:16 +0100357 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400358 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400359
Peter Staubach38c73042009-08-10 08:54:16 -0400360start:
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400361 /*
362 * Prevent starvation issues if someone is doing a consistency
363 * sync-to-disk
364 */
365 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
366 nfs_wait_bit_killable, TASK_KILLABLE);
367 if (ret)
368 return ret;
369
Nick Piggin54566b22009-01-04 12:00:53 -0800370 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700371 if (!page)
372 return -ENOMEM;
373 *pagep = page;
374
375 ret = nfs_flush_incompatible(file, page);
376 if (ret) {
377 unlock_page(page);
378 page_cache_release(page);
Peter Staubach38c73042009-08-10 08:54:16 -0400379 } else if (!once_thru &&
380 nfs_want_read_modify_write(file, page, pos, len)) {
381 once_thru = 1;
382 ret = nfs_readpage(file, page);
383 page_cache_release(page);
384 if (!ret)
385 goto start;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700386 }
387 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389
Nick Piggin4899f9c2007-10-16 01:25:16 -0700390static int nfs_write_end(struct file *file, struct address_space *mapping,
391 loff_t pos, unsigned len, unsigned copied,
392 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700394 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
Andy Adamsondc248262013-08-14 11:59:16 -0400395 struct nfs_open_context *ctx = nfs_file_open_context(file);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700396 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Niels de Vos1e8968c2013-12-17 18:20:16 +0100398 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400399 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400400
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400401 /*
402 * Zero any uninitialised parts of the page, and then mark the page
403 * as up to date if it turns out that we're extending the file.
404 */
405 if (!PageUptodate(page)) {
406 unsigned pglen = nfs_page_length(page);
407 unsigned end = offset + len;
408
409 if (pglen == 0) {
410 zero_user_segments(page, 0, offset,
411 end, PAGE_CACHE_SIZE);
412 SetPageUptodate(page);
413 } else if (end >= pglen) {
414 zero_user_segment(page, end, PAGE_CACHE_SIZE);
415 if (offset == 0)
416 SetPageUptodate(page);
417 } else
418 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
419 }
420
Nick Piggin4899f9c2007-10-16 01:25:16 -0700421 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700422
423 unlock_page(page);
424 page_cache_release(page);
425
Chuck Lever3d509e52007-12-20 14:55:04 -0500426 if (status < 0)
427 return status;
Andy Adamson2701d082012-05-24 13:13:24 -0400428 NFS_I(mapping->host)->write_io += copied;
Andy Adamsondc248262013-08-14 11:59:16 -0400429
430 if (nfs_ctx_key_to_expire(ctx)) {
431 status = nfs_wb_all(mapping->host);
432 if (status < 0)
433 return status;
434 }
435
Chuck Lever3d509e52007-12-20 14:55:04 -0500436 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
David Howells6b9b3512009-04-03 16:42:41 +0100439/*
440 * Partially or wholly invalidate a page
441 * - Release the private state associated with a page if undergoing complete
442 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100443 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100444 * - Caller holds page lock
445 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400446static void nfs_invalidate_page(struct page *page, unsigned int offset,
447 unsigned int length)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500448{
Lukas Czernerd47992f2013-05-21 23:17:23 -0400449 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
450 page, offset, length);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400451
Lukas Czernerd47992f2013-05-21 23:17:23 -0400452 if (offset != 0 || length < PAGE_CACHE_SIZE)
Trond Myklebust1c759502006-10-09 16:18:38 -0400453 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400454 /* Cancel any unstarted writes on this page */
Mel Gormand56b4dd2012-07-31 16:45:06 -0700455 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
David Howells545db452009-04-03 16:42:44 +0100456
457 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500458}
459
David Howells6b9b3512009-04-03 16:42:41 +0100460/*
461 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100462 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100463 * - Caller holds page lock
464 * - Return true (may release page) or false (may not)
465 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500466static int nfs_release_page(struct page *page, gfp_t gfp)
467{
Trond Myklebustb608b282010-07-30 15:31:54 -0400468 struct address_space *mapping = page->mapping;
469
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400470 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
471
Jeff Layton5cf02d02012-07-23 13:58:51 -0400472 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
473 * doing this memory reclaim for a fs-related allocation.
474 */
475 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
476 !(current->flags & PF_FSTRANS)) {
Trond Myklebustb608b282010-07-30 15:31:54 -0400477 int how = FLUSH_SYNC;
478
479 /* Don't let kswapd deadlock waiting for OOM RPC calls */
480 if (current_is_kswapd())
481 how = 0;
482 nfs_commit_inode(mapping->host, how);
483 }
Trond Myklebuste3db7692007-01-10 23:15:39 -0800484 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100485 if (PagePrivate(page))
486 return 0;
487 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800488}
489
Mel Gormanf919b192013-07-03 15:02:06 -0700490static void nfs_check_dirty_writeback(struct page *page,
491 bool *dirty, bool *writeback)
492{
493 struct nfs_inode *nfsi;
494 struct address_space *mapping = page_file_mapping(page);
495
496 if (!mapping || PageSwapCache(page))
497 return;
498
499 /*
500 * Check if an unstable page is currently being committed and
501 * if so, have the VM treat it as if the page is under writeback
502 * so it will not block due to pages that will shortly be freeable.
503 */
504 nfsi = NFS_I(mapping->host);
505 if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
506 *writeback = true;
507 return;
508 }
509
510 /*
511 * If PagePrivate() is set, then the page is not freeable and as the
512 * inode is not being committed, it's not going to be cleaned in the
513 * near future so treat it as dirty
514 */
515 if (PagePrivate(page))
516 *dirty = true;
517}
518
David Howells6b9b3512009-04-03 16:42:41 +0100519/*
520 * Attempt to clear the private state associated with a page when an error
521 * occurs that requires the cached contents of an inode to be written back or
522 * destroyed
David Howells545db452009-04-03 16:42:44 +0100523 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100524 * - Caller holds page lock
525 * - Return 0 if successful, -error otherwise
526 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800527static int nfs_launder_page(struct page *page)
528{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700529 struct inode *inode = page_file_mapping(page)->host;
David Howells545db452009-04-03 16:42:44 +0100530 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400531
532 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
533 inode->i_ino, (long long)page_offset(page));
534
David Howells545db452009-04-03 16:42:44 +0100535 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400536 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500537}
538
Mel Gormana564b8f2012-07-31 16:45:12 -0700539#ifdef CONFIG_NFS_SWAP
540static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
541 sector_t *span)
542{
543 *span = sis->pages;
544 return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
545}
546
547static void nfs_swap_deactivate(struct file *file)
548{
549 xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
550}
551#endif
552
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700553const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 .readpage = nfs_readpage,
555 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400556 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 .writepage = nfs_writepage,
558 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700559 .write_begin = nfs_write_begin,
560 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500561 .invalidatepage = nfs_invalidate_page,
562 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 .direct_IO = nfs_direct_IO,
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400564 .migratepage = nfs_migrate_page,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800565 .launder_page = nfs_launder_page,
Mel Gormanf919b192013-07-03 15:02:06 -0700566 .is_dirty_writeback = nfs_check_dirty_writeback,
Andi Kleenf590f332009-09-16 11:50:17 +0200567 .error_remove_page = generic_error_remove_page,
Mel Gormana564b8f2012-07-31 16:45:12 -0700568#ifdef CONFIG_NFS_SWAP
569 .swap_activate = nfs_swap_activate,
570 .swap_deactivate = nfs_swap_deactivate,
571#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572};
573
David Howells6b9b3512009-04-03 16:42:41 +0100574/*
575 * Notification that a PTE pointing to an NFS page is about to be made
576 * writable, implying that someone is about to modify the page through a
577 * shared-writable mapping
578 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700579static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400580{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700581 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400582 struct file *filp = vma->vm_file;
Al Viro6de14722013-09-16 10:53:17 -0400583 struct inode *inode = file_inode(filp);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400584 unsigned pagelen;
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400585 int ret = VM_FAULT_NOPAGE;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700586 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400587
Niels de Vos1e8968c2013-12-17 18:20:16 +0100588 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400589 filp, filp->f_mapping->host->i_ino,
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400590 (long long)page_offset(page));
591
David Howells545db452009-04-03 16:42:44 +0100592 /* make sure the cache has finished storing the page */
Al Viro6de14722013-09-16 10:53:17 -0400593 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
David Howells545db452009-04-03 16:42:44 +0100594
Trond Myklebust94387fb2007-07-22 17:09:05 -0400595 lock_page(page);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700596 mapping = page_file_mapping(page);
Al Viro6de14722013-09-16 10:53:17 -0400597 if (mapping != inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500598 goto out_unlock;
599
Trond Myklebust2aeb98f2012-01-17 22:04:26 -0500600 wait_on_page_writeback(page);
601
Trond Myklebust94387fb2007-07-22 17:09:05 -0400602 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500603 if (pagelen == 0)
604 goto out_unlock;
605
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400606 ret = VM_FAULT_LOCKED;
607 if (nfs_flush_incompatible(filp, page) == 0 &&
608 nfs_updatepage(filp, page, 0, pagelen) == 0)
609 goto out;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500610
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400611 ret = VM_FAULT_SIGBUS;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500612out_unlock:
Trond Myklebust94387fb2007-07-22 17:09:05 -0400613 unlock_page(page);
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400614out:
615 return ret;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400616}
617
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400618static const struct vm_operations_struct nfs_file_vm_ops = {
Trond Myklebust94387fb2007-07-22 17:09:05 -0400619 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700620 .map_pages = filemap_map_pages,
Trond Myklebust94387fb2007-07-22 17:09:05 -0400621 .page_mkwrite = nfs_vm_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700622 .remap_pages = generic_file_remap_pages,
Trond Myklebust94387fb2007-07-22 17:09:05 -0400623};
624
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400625static int nfs_need_sync_write(struct file *filp, struct inode *inode)
626{
627 struct nfs_open_context *ctx;
628
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100629 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400630 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400631 ctx = nfs_file_open_context(filp);
Andy Adamsondc248262013-08-14 11:59:16 -0400632 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
633 nfs_ctx_key_to_expire(ctx))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400634 return 1;
635 return 0;
636}
637
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400638ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
639 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Al Viro6de14722013-09-16 10:53:17 -0400641 struct file *file = iocb->ki_filp;
642 struct inode *inode = file_inode(file);
Chuck Lever7e381172010-02-01 14:17:41 -0500643 unsigned long written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700645 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Al Viro6de14722013-09-16 10:53:17 -0400647 result = nfs_key_timeout_notify(file, inode);
Andy Adamsondc248262013-08-14 11:59:16 -0400648 if (result)
649 return result;
650
Al Viro6de14722013-09-16 10:53:17 -0400651 if (file->f_flags & O_DIRECT)
Mel Gormana564b8f2012-07-31 16:45:12 -0700652 return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Al Viro6de14722013-09-16 10:53:17 -0400654 dprintk("NFS: write(%pD2, %lu@%Ld)\n",
655 file, (unsigned long) count, (long long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 result = -EBUSY;
658 if (IS_SWAPFILE(inode))
659 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000660 /*
661 * O_APPEND implies that we must revalidate the file length.
662 */
Al Viro6de14722013-09-16 10:53:17 -0400663 if (file->f_flags & O_APPEND) {
664 result = nfs_revalidate_file_size(inode, file);
Trond Myklebust7d52e862005-06-22 17:16:30 +0000665 if (result)
666 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
669 result = count;
670 if (!count)
671 goto out;
672
Badari Pulavarty027445c2006-09-30 23:28:46 -0700673 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
Chuck Lever7e381172010-02-01 14:17:41 -0500674 if (result > 0)
675 written = result;
676
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100677 /* Return error values for O_DSYNC and IS_SYNC() */
Al Viro6de14722013-09-16 10:53:17 -0400678 if (result >= 0 && nfs_need_sync_write(file, inode)) {
679 int err = vfs_fsync(file, 0);
Trond Myklebust200baa22006-12-05 00:35:40 -0500680 if (err < 0)
681 result = err;
682 }
Chuck Lever7e381172010-02-01 14:17:41 -0500683 if (result > 0)
684 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685out:
686 return result;
687
688out_swapfile:
689 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
690 goto out;
691}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400692EXPORT_SYMBOL_GPL(nfs_file_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400694ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
695 struct file *filp, loff_t *ppos,
696 size_t count, unsigned int flags)
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700697{
Al Viro6de14722013-09-16 10:53:17 -0400698 struct inode *inode = file_inode(filp);
Chuck Lever7e381172010-02-01 14:17:41 -0500699 unsigned long written = 0;
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700700 ssize_t ret;
701
Al Viro6de14722013-09-16 10:53:17 -0400702 dprintk("NFS splice_write(%pD2, %lu@%llu)\n",
703 filp, (unsigned long) count, (unsigned long long) *ppos);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700704
705 /*
706 * The combination of splice and an O_APPEND destination is disallowed.
707 */
708
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700709 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
Chuck Lever7e381172010-02-01 14:17:41 -0500710 if (ret > 0)
711 written = ret;
712
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700713 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400714 int err = vfs_fsync(filp, 0);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700715 if (err < 0)
716 ret = err;
717 }
Chuck Lever7e381172010-02-01 14:17:41 -0500718 if (ret > 0)
719 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700720 return ret;
721}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400722EXPORT_SYMBOL_GPL(nfs_file_splice_write);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700723
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400724static int
725do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
727 struct inode *inode = filp->f_mapping->host;
728 int status = 0;
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000729 unsigned int saved_type = fl->fl_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Trond Myklebust039c4d72005-10-18 14:20:16 -0700731 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400732 posix_test_lock(filp, fl);
733 if (fl->fl_type != F_UNLCK) {
734 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700735 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000737 fl->fl_type = saved_type;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700738
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400739 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
Trond Myklebust039c4d72005-10-18 14:20:16 -0700740 goto out_noconflict;
741
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400742 if (is_local)
Trond Myklebust039c4d72005-10-18 14:20:16 -0700743 goto out_noconflict;
744
745 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
746out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700748out_noconflict:
749 fl->fl_type = F_UNLCK;
750 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
753static int do_vfs_lock(struct file *file, struct file_lock *fl)
754{
755 int res = 0;
756 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
757 case FL_POSIX:
758 res = posix_lock_file_wait(file, fl);
759 break;
760 case FL_FLOCK:
761 res = flock_lock_file_wait(file, fl);
762 break;
763 default:
764 BUG();
765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 return res;
767}
768
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400769static int
770do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
772 struct inode *inode = filp->f_mapping->host;
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400773 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 int status;
775
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 /*
777 * Flush all pending writes before doing anything
778 * with locks..
779 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500780 nfs_sync_mapping(filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400782 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
783 if (!IS_ERR(l_ctx)) {
784 status = nfs_iocounter_wait(&l_ctx->io_count);
785 nfs_put_lock_context(l_ctx);
786 if (status < 0)
787 return status;
788 }
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /* NOTE: special case
791 * If we're signalled while cleaning up locks on process exit, we
792 * still need to complete the unlock.
793 */
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400794 /*
795 * Use local locking if mounted with "-onolock" or with appropriate
796 * "-olocal_lock="
797 */
798 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
800 else
801 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 return status;
803}
804
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400805static int
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700806is_time_granular(struct timespec *ts) {
807 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
808}
809
810static int
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400811do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
813 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 int status;
815
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 /*
817 * Flush all pending writes before doing anything
818 * with locks..
819 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500820 status = nfs_sync_mapping(filp->f_mapping);
821 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 goto out;
823
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400824 /*
825 * Use local locking if mounted with "-onolock" or with appropriate
826 * "-olocal_lock="
827 */
828 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400830 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 if (status < 0)
833 goto out;
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 /*
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700836 * Revalidate the cache if the server has time stamps granular
837 * enough to detect subsecond changes. Otherwise, clear the
838 * cache to prevent missing any changes.
839 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 * This makes locking act as a cache coherency point.
841 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500842 nfs_sync_mapping(filp->f_mapping);
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400843 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700844 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
845 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
846 else
847 nfs_zap_caches(inode);
848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 return status;
851}
852
853/*
854 * Lock a (portion of) a file
855 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400856int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400858 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400859 int ret = -ENOLCK;
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400860 int is_local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
Al Viro6de14722013-09-16 10:53:17 -0400862 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
863 filp, fl->fl_type, fl->fl_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400865
Chuck Lever91d5b472006-03-20 13:44:14 -0500866 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700869 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400870 goto out_err;
871
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400872 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
873 is_local = 1;
874
Trond Myklebust21162712008-05-20 19:34:39 -0400875 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
876 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
877 if (ret < 0)
878 goto out_err;
879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 if (IS_GETLK(cmd))
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400882 ret = do_getlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400883 else if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400884 ret = do_unlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400885 else
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400886 ret = do_setlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400887out_err:
888 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400890EXPORT_SYMBOL_GPL(nfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892/*
893 * Lock a (portion of) a file
894 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400895int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400897 struct inode *inode = filp->f_mapping->host;
898 int is_local = 0;
899
Al Viro6de14722013-09-16 10:53:17 -0400900 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
901 filp, fl->fl_type, fl->fl_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (!(fl->fl_flags & FL_FLOCK))
904 return -ENOLCK;
905
Jeff Laytonad0fcd42012-07-23 15:46:23 -0400906 /*
907 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
908 * any standard. In principle we might be able to support LOCK_MAND
909 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
910 * NFS code is not set up for it.
911 */
912 if (fl->fl_type & LOCK_MAND)
913 return -EINVAL;
914
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400915 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
916 is_local = 1;
917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 /* We're simulating flock() locks using posix locks on the server */
919 fl->fl_owner = (fl_owner_t)filp;
920 fl->fl_start = 0;
921 fl->fl_end = OFFSET_MAX;
922
923 if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400924 return do_unlk(filp, cmd, fl, is_local);
925 return do_setlk(filp, cmd, fl, is_local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400927EXPORT_SYMBOL_GPL(nfs_flock);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400928
Chuck Lever6da24bc2008-06-11 17:55:58 -0400929/*
930 * There is no protocol support for leases, so we have no way to implement
931 * them correctly in the face of opens by other clients.
932 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400933int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
J. Bruce Fields370f6592007-06-08 15:23:34 -0400934{
Al Viro6de14722013-09-16 10:53:17 -0400935 dprintk("NFS: setlease(%pD2, arg=%ld)\n", file, arg);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400936 return -EINVAL;
937}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400938EXPORT_SYMBOL_GPL(nfs_setlease);
Jeff Layton1788ea62011-11-04 13:31:21 -0400939
Jeff Layton04869582011-11-04 13:31:22 -0400940const struct file_operations nfs_file_operations = {
941 .llseek = nfs_file_llseek,
942 .read = do_sync_read,
943 .write = do_sync_write,
944 .aio_read = nfs_file_read,
945 .aio_write = nfs_file_write,
946 .mmap = nfs_file_mmap,
947 .open = nfs_file_open,
948 .flush = nfs_file_flush,
949 .release = nfs_file_release,
950 .fsync = nfs_file_fsync,
951 .lock = nfs_lock,
952 .flock = nfs_flock,
953 .splice_read = nfs_file_splice_read,
954 .splice_write = nfs_file_splice_write,
955 .check_flags = nfs_check_flags,
956 .setlease = nfs_setlease,
957};
Bryan Schumakerddda8e02012-07-30 16:05:23 -0400958EXPORT_SYMBOL_GPL(nfs_file_operations);