blob: a87a44f8411304218986faa171f40de1f9564cd2 [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
40#define NFSDBG_FACILITY NFSDBG_FILE
41
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040042static const struct vm_operations_struct nfs_file_vm_ops;
Trond Myklebust94387fb2007-07-22 17:09:05 -040043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/* Hack for future NFS swap support */
45#ifndef IS_SWAPFILE
46# define IS_SWAPFILE(inode) (0)
47#endif
48
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -040049int nfs_check_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070050{
51 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
52 return -EINVAL;
53
54 return 0;
55}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040056EXPORT_SYMBOL_GPL(nfs_check_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58/*
59 * Open file
60 */
61static int
62nfs_file_open(struct inode *inode, struct file *filp)
63{
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 int res;
65
Chuck Lever6da24bc2008-06-11 17:55:58 -040066 dprintk("NFS: open file(%s/%s)\n",
Chuck Levercc0dd2d2008-06-11 17:55:42 -040067 filp->f_path.dentry->d_parent->d_name.name,
68 filp->f_path.dentry->d_name.name);
69
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{
Chuck Lever6da24bc2008-06-11 17:55:58 -040082 dprintk("NFS: release(%s/%s)\n",
Rakib Mullick6f276e42011-11-01 12:16:15 +060083 filp->f_path.dentry->d_parent->d_name.name,
84 filp->f_path.dentry->d_name.name);
Chuck Lever6da24bc2008-06-11 17:55:58 -040085
Chuck Lever91d5b472006-03-20 13:44:14 -050086 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Trond Myklebust46cb6502008-06-11 16:32:46 -040087 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040089EXPORT_SYMBOL_GPL(nfs_file_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Trond Myklebust980802e2005-06-13 11:14:01 -040091/**
92 * nfs_revalidate_size - Revalidate the file size
93 * @inode - pointer to inode struct
94 * @file - pointer to struct file
95 *
96 * Revalidates the file length. This is basically a wrapper around
97 * nfs_revalidate_inode() that takes into account the fact that we may
98 * have cached writes (in which case we don't care about the server's
99 * idea of what the file length is), or O_DIRECT (in which case we
100 * shouldn't trust the cache).
101 */
102static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
103{
104 struct nfs_server *server = NFS_SERVER(inode);
105 struct nfs_inode *nfsi = NFS_I(inode);
106
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400107 if (nfs_have_delegated_attributes(inode))
108 goto out_noreval;
109
Trond Myklebust980802e2005-06-13 11:14:01 -0400110 if (filp->f_flags & O_DIRECT)
111 goto force_reval;
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400112 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
113 goto force_reval;
114 if (nfs_attribute_timeout(inode))
115 goto force_reval;
116out_noreval:
117 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400118force_reval:
119 return __nfs_revalidate_inode(server, inode);
120}
121
Andrew Morton965c8e52012-12-17 15:59:39 -0800122loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
Trond Myklebust980802e2005-06-13 11:14:01 -0400123{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400124 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
Chuck Leverb84e06c2008-06-11 17:55:34 -0400125 filp->f_path.dentry->d_parent->d_name.name,
126 filp->f_path.dentry->d_name.name,
Andrew Morton965c8e52012-12-17 15:59:39 -0800127 offset, whence);
Chuck Leverb84e06c2008-06-11 17:55:34 -0400128
Josef Bacik06222e42011-07-18 13:21:38 -0400129 /*
Andrew Morton965c8e52012-12-17 15:59:39 -0800130 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
Josef Bacik06222e42011-07-18 13:21:38 -0400131 * the cached file length
132 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800133 if (whence != SEEK_SET && whence != SEEK_CUR) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400134 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400135
Trond Myklebust980802e2005-06-13 11:14:01 -0400136 int retval = nfs_revalidate_file_size(inode, filp);
137 if (retval < 0)
138 return (loff_t)retval;
Andi Kleen79835a712011-09-15 16:06:52 -0700139 }
Trond Myklebustd5e66342008-09-23 17:28:35 -0400140
Andrew Morton965c8e52012-12-17 15:59:39 -0800141 return generic_file_llseek(filp, offset, whence);
Trond Myklebust980802e2005-06-13 11:14:01 -0400142}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400143EXPORT_SYMBOL_GPL(nfs_file_llseek);
Trond Myklebust980802e2005-06-13 11:14:01 -0400144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145/*
146 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400148int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700149nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400151 struct dentry *dentry = file->f_path.dentry;
152 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Chuck Lever6da24bc2008-06-11 17:55:58 -0400154 dprintk("NFS: flush(%s/%s)\n",
155 dentry->d_parent->d_name.name,
156 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Chuck Leverc2459dc2010-02-01 14:17:14 -0500158 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 if ((file->f_mode & FMODE_WRITE) == 0)
160 return 0;
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400161
Trond Myklebust14546c32012-05-06 19:10:59 -0400162 /*
163 * If we're holding a write delegation, then just start the i/o
164 * but don't wait for completion (or send a commit).
165 */
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400166 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebust14546c32012-05-06 19:10:59 -0400167 return filemap_fdatawrite(file->f_mapping);
168
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400169 /* Flush writes to the server and return any errors */
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400170 return vfs_fsync(file, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400172EXPORT_SYMBOL_GPL(nfs_file_flush);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400174ssize_t
Badari Pulavarty027445c2006-09-30 23:28:46 -0700175nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
176 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800178 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 struct inode * inode = dentry->d_inode;
180 ssize_t result;
181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 if (iocb->ki_filp->f_flags & O_DIRECT)
Mel Gormana564b8f2012-07-31 16:45:12 -0700183 return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Chuck Lever6da24bc2008-06-11 17:55:58 -0400185 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 dentry->d_parent->d_name.name, dentry->d_name.name,
Rakib Mullick6f276e42011-11-01 12:16:15 +0600187 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Trond Myklebust44b11872006-05-25 01:40:59 -0400189 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500190 if (!result) {
Badari Pulavarty027445c2006-09-30 23:28:46 -0700191 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500192 if (result > 0)
193 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 return result;
196}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400197EXPORT_SYMBOL_GPL(nfs_file_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400199ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200200nfs_file_splice_read(struct file *filp, loff_t *ppos,
201 struct pipe_inode_info *pipe, size_t count,
202 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800204 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 struct inode *inode = dentry->d_inode;
206 ssize_t res;
207
Chuck Lever6da24bc2008-06-11 17:55:58 -0400208 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 dentry->d_parent->d_name.name, dentry->d_name.name,
210 (unsigned long) count, (unsigned long long) *ppos);
211
Trond Myklebust44b11872006-05-25 01:40:59 -0400212 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500213 if (!res) {
Jens Axboef0930ff2007-06-01 11:51:43 +0200214 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500215 if (res > 0)
216 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return res;
219}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400220EXPORT_SYMBOL_GPL(nfs_file_splice_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400222int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
224{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800225 struct dentry *dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 struct inode *inode = dentry->d_inode;
227 int status;
228
Chuck Lever6da24bc2008-06-11 17:55:58 -0400229 dprintk("NFS: mmap(%s/%s)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 dentry->d_parent->d_name.name, dentry->d_name.name);
231
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400232 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
233 * so we call that before revalidating the mapping
234 */
235 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400236 if (!status) {
237 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400238 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400239 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 return status;
241}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400242EXPORT_SYMBOL_GPL(nfs_file_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244/*
245 * Flush any dirty pages for this process, and check for write errors.
246 * The return status from this call provides a reliable indication of
247 * whether any write errors occurred for this process.
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400248 *
249 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
250 * disk, but it retrieves and clears ctx->error after synching, despite
251 * the two being set at the same time in nfs_context_set_write_error().
252 * This is because the former is used to notify the _next_ call to
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300253 * nfs_file_write() that a write error occurred, and hence cause it to
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400254 * fall back to doing a synchronous write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400256int
Bryan Schumakera5c58892012-06-20 15:53:42 -0400257nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200259 struct dentry *dentry = file->f_path.dentry;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400260 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 struct inode *inode = dentry->d_inode;
Trond Myklebust05990d12012-09-11 16:01:22 -0400262 int have_error, do_resend, status;
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400263 int ret = 0;
264
Chuck Lever6da24bc2008-06-11 17:55:58 -0400265 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
Chuck Lever54917782008-05-27 16:29:07 -0400266 dentry->d_parent->d_name.name, dentry->d_name.name,
267 datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Chuck Lever91d5b472006-03-20 13:44:14 -0500269 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebust05990d12012-09-11 16:01:22 -0400270 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400271 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
272 status = nfs_commit_inode(inode, FLUSH_SYNC);
273 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
Trond Myklebust05990d12012-09-11 16:01:22 -0400274 if (have_error) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400275 ret = xchg(&ctx->error, 0);
Trond Myklebust05990d12012-09-11 16:01:22 -0400276 if (ret)
277 goto out;
278 }
279 if (status < 0) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400280 ret = status;
Trond Myklebust05990d12012-09-11 16:01:22 -0400281 goto out;
282 }
283 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
284 if (do_resend)
285 ret = -EAGAIN;
286out:
Bryan Schumakera5c58892012-06-20 15:53:42 -0400287 return ret;
288}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400289EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
Bryan Schumakera5c58892012-06-20 15:53:42 -0400290
291static int
292nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
293{
294 int ret;
Al Viro496ad9a2013-01-23 17:07:38 -0500295 struct inode *inode = file_inode(file);
Bryan Schumakera5c58892012-06-20 15:53:42 -0400296
Trond Myklebust05990d12012-09-11 16:01:22 -0400297 do {
298 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
299 if (ret != 0)
300 break;
301 mutex_lock(&inode->i_mutex);
302 ret = nfs_file_fsync_commit(file, start, end, datasync);
303 mutex_unlock(&inode->i_mutex);
Trond Myklebustdcfc4f22012-09-11 16:19:38 -0400304 /*
305 * If nfs_file_fsync_commit detected a server reboot, then
306 * resend all dirty pages that might have been covered by
307 * the NFS_CONTEXT_RESEND_WRITES flag
308 */
309 start = 0;
310 end = LLONG_MAX;
Trond Myklebust05990d12012-09-11 16:01:22 -0400311 } while (ret == -EAGAIN);
312
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400313 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314}
315
316/*
Peter Staubach38c73042009-08-10 08:54:16 -0400317 * Decide whether a read/modify/write cycle may be more efficient
318 * then a modify/write/read cycle when writing to a page in the
319 * page cache.
320 *
321 * The modify/write/read cycle may occur if a page is read before
322 * being completely filled by the writer. In this situation, the
323 * page must be completely written to stable storage on the server
324 * before it can be refilled by reading in the page from the server.
325 * This can lead to expensive, small, FILE_SYNC mode writes being
326 * done.
327 *
328 * It may be more efficient to read the page first if the file is
329 * open for reading in addition to writing, the page is not marked
330 * as Uptodate, it is not dirty or waiting to be committed,
331 * indicating that it was previously allocated and then modified,
332 * that there were valid bytes of data in that range of the file,
333 * and that the new data won't completely replace the old data in
334 * that range of the file.
335 */
336static int nfs_want_read_modify_write(struct file *file, struct page *page,
337 loff_t pos, unsigned len)
338{
339 unsigned int pglen = nfs_page_length(page);
340 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
341 unsigned int end = offset + len;
342
343 if ((file->f_mode & FMODE_READ) && /* open for read? */
344 !PageUptodate(page) && /* Uptodate? */
345 !PagePrivate(page) && /* i/o request already? */
346 pglen && /* valid bytes of file? */
347 (end < pglen || offset)) /* replace all valid bytes? */
348 return 1;
349 return 0;
350}
351
352/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700353 * This does the "real" work of the write. We must allocate and lock the
354 * page to be sent back to the generic routine, which then copies the
355 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 *
357 * If the writer ends up delaying the write, the writer needs to
358 * increment the page use counts until he is done with the page.
359 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700360static int nfs_write_begin(struct file *file, struct address_space *mapping,
361 loff_t pos, unsigned len, unsigned flags,
362 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700364 int ret;
Peter Staubach38c73042009-08-10 08:54:16 -0400365 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700366 struct page *page;
Peter Staubach38c73042009-08-10 08:54:16 -0400367 int once_thru = 0;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700368
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400369 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
370 file->f_path.dentry->d_parent->d_name.name,
371 file->f_path.dentry->d_name.name,
372 mapping->host->i_ino, len, (long long) pos);
373
Peter Staubach38c73042009-08-10 08:54:16 -0400374start:
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400375 /*
376 * Prevent starvation issues if someone is doing a consistency
377 * sync-to-disk
378 */
379 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
380 nfs_wait_bit_killable, TASK_KILLABLE);
381 if (ret)
382 return ret;
383
Nick Piggin54566b22009-01-04 12:00:53 -0800384 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700385 if (!page)
386 return -ENOMEM;
387 *pagep = page;
388
389 ret = nfs_flush_incompatible(file, page);
390 if (ret) {
391 unlock_page(page);
392 page_cache_release(page);
Peter Staubach38c73042009-08-10 08:54:16 -0400393 } else if (!once_thru &&
394 nfs_want_read_modify_write(file, page, pos, len)) {
395 once_thru = 1;
396 ret = nfs_readpage(file, page);
397 page_cache_release(page);
398 if (!ret)
399 goto start;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700400 }
401 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Nick Piggin4899f9c2007-10-16 01:25:16 -0700404static int nfs_write_end(struct file *file, struct address_space *mapping,
405 loff_t pos, unsigned len, unsigned copied,
406 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700408 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
409 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400411 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
412 file->f_path.dentry->d_parent->d_name.name,
413 file->f_path.dentry->d_name.name,
414 mapping->host->i_ino, len, (long long) pos);
415
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400416 /*
417 * Zero any uninitialised parts of the page, and then mark the page
418 * as up to date if it turns out that we're extending the file.
419 */
420 if (!PageUptodate(page)) {
421 unsigned pglen = nfs_page_length(page);
422 unsigned end = offset + len;
423
424 if (pglen == 0) {
425 zero_user_segments(page, 0, offset,
426 end, PAGE_CACHE_SIZE);
427 SetPageUptodate(page);
428 } else if (end >= pglen) {
429 zero_user_segment(page, end, PAGE_CACHE_SIZE);
430 if (offset == 0)
431 SetPageUptodate(page);
432 } else
433 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
434 }
435
Nick Piggin4899f9c2007-10-16 01:25:16 -0700436 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700437
438 unlock_page(page);
439 page_cache_release(page);
440
Chuck Lever3d509e52007-12-20 14:55:04 -0500441 if (status < 0)
442 return status;
Andy Adamson2701d082012-05-24 13:13:24 -0400443 NFS_I(mapping->host)->write_io += copied;
Chuck Lever3d509e52007-12-20 14:55:04 -0500444 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
446
David Howells6b9b3512009-04-03 16:42:41 +0100447/*
448 * Partially or wholly invalidate a page
449 * - Release the private state associated with a page if undergoing complete
450 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100451 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100452 * - Caller holds page lock
453 */
NeilBrown2ff28e22006-03-26 01:37:18 -0800454static void nfs_invalidate_page(struct page *page, unsigned long offset)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500455{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400456 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
457
Trond Myklebust1c759502006-10-09 16:18:38 -0400458 if (offset != 0)
459 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
Jeff Layton5cf02d02012-07-23 13:58:51 -0400478 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
479 * doing this memory reclaim for a fs-related allocation.
480 */
481 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
482 !(current->flags & PF_FSTRANS)) {
Trond Myklebustb608b282010-07-30 15:31:54 -0400483 int how = FLUSH_SYNC;
484
485 /* Don't let kswapd deadlock waiting for OOM RPC calls */
486 if (current_is_kswapd())
487 how = 0;
488 nfs_commit_inode(mapping->host, how);
489 }
Trond Myklebuste3db7692007-01-10 23:15:39 -0800490 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100491 if (PagePrivate(page))
492 return 0;
493 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800494}
495
David Howells6b9b3512009-04-03 16:42:41 +0100496/*
497 * Attempt to clear the private state associated with a page when an error
498 * occurs that requires the cached contents of an inode to be written back or
499 * destroyed
David Howells545db452009-04-03 16:42:44 +0100500 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100501 * - Caller holds page lock
502 * - Return 0 if successful, -error otherwise
503 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800504static int nfs_launder_page(struct page *page)
505{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700506 struct inode *inode = page_file_mapping(page)->host;
David Howells545db452009-04-03 16:42:44 +0100507 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400508
509 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
510 inode->i_ino, (long long)page_offset(page));
511
David Howells545db452009-04-03 16:42:44 +0100512 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400513 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500514}
515
Mel Gormana564b8f2012-07-31 16:45:12 -0700516#ifdef CONFIG_NFS_SWAP
517static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
518 sector_t *span)
519{
520 *span = sis->pages;
521 return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
522}
523
524static void nfs_swap_deactivate(struct file *file)
525{
526 xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
527}
528#endif
529
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700530const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 .readpage = nfs_readpage,
532 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400533 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 .writepage = nfs_writepage,
535 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700536 .write_begin = nfs_write_begin,
537 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500538 .invalidatepage = nfs_invalidate_page,
539 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 .direct_IO = nfs_direct_IO,
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400541 .migratepage = nfs_migrate_page,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800542 .launder_page = nfs_launder_page,
Andi Kleenf590f332009-09-16 11:50:17 +0200543 .error_remove_page = generic_error_remove_page,
Mel Gormana564b8f2012-07-31 16:45:12 -0700544#ifdef CONFIG_NFS_SWAP
545 .swap_activate = nfs_swap_activate,
546 .swap_deactivate = nfs_swap_deactivate,
547#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548};
549
David Howells6b9b3512009-04-03 16:42:41 +0100550/*
551 * Notification that a PTE pointing to an NFS page is about to be made
552 * writable, implying that someone is about to modify the page through a
553 * shared-writable mapping
554 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700555static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400556{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700557 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400558 struct file *filp = vma->vm_file;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400559 struct dentry *dentry = filp->f_path.dentry;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400560 unsigned pagelen;
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400561 int ret = VM_FAULT_NOPAGE;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700562 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400563
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400564 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
565 dentry->d_parent->d_name.name, dentry->d_name.name,
566 filp->f_mapping->host->i_ino,
567 (long long)page_offset(page));
568
David Howells545db452009-04-03 16:42:44 +0100569 /* make sure the cache has finished storing the page */
570 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
571
Trond Myklebust94387fb2007-07-22 17:09:05 -0400572 lock_page(page);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700573 mapping = page_file_mapping(page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400574 if (mapping != dentry->d_inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500575 goto out_unlock;
576
Trond Myklebust2aeb98f2012-01-17 22:04:26 -0500577 wait_on_page_writeback(page);
578
Trond Myklebust94387fb2007-07-22 17:09:05 -0400579 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500580 if (pagelen == 0)
581 goto out_unlock;
582
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400583 ret = VM_FAULT_LOCKED;
584 if (nfs_flush_incompatible(filp, page) == 0 &&
585 nfs_updatepage(filp, page, 0, pagelen) == 0)
586 goto out;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500587
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400588 ret = VM_FAULT_SIGBUS;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500589out_unlock:
Trond Myklebust94387fb2007-07-22 17:09:05 -0400590 unlock_page(page);
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400591out:
592 return ret;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400593}
594
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400595static const struct vm_operations_struct nfs_file_vm_ops = {
Trond Myklebust94387fb2007-07-22 17:09:05 -0400596 .fault = filemap_fault,
597 .page_mkwrite = nfs_vm_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700598 .remap_pages = generic_file_remap_pages,
Trond Myklebust94387fb2007-07-22 17:09:05 -0400599};
600
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400601static int nfs_need_sync_write(struct file *filp, struct inode *inode)
602{
603 struct nfs_open_context *ctx;
604
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100605 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400606 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400607 ctx = nfs_file_open_context(filp);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400608 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
609 return 1;
610 return 0;
611}
612
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400613ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
614 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800616 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 struct inode * inode = dentry->d_inode;
Chuck Lever7e381172010-02-01 14:17:41 -0500618 unsigned long written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700620 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (iocb->ki_filp->f_flags & O_DIRECT)
Mel Gormana564b8f2012-07-31 16:45:12 -0700623 return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Chuck Lever6da24bc2008-06-11 17:55:58 -0400625 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 dentry->d_parent->d_name.name, dentry->d_name.name,
Chuck Lever6da24bc2008-06-11 17:55:58 -0400627 (unsigned long) count, (long long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 result = -EBUSY;
630 if (IS_SWAPFILE(inode))
631 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000632 /*
633 * O_APPEND implies that we must revalidate the file length.
634 */
635 if (iocb->ki_filp->f_flags & O_APPEND) {
636 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
637 if (result)
638 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000639 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
641 result = count;
642 if (!count)
643 goto out;
644
Badari Pulavarty027445c2006-09-30 23:28:46 -0700645 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
Chuck Lever7e381172010-02-01 14:17:41 -0500646 if (result > 0)
647 written = result;
648
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100649 /* Return error values for O_DSYNC and IS_SYNC() */
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400650 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400651 int err = vfs_fsync(iocb->ki_filp, 0);
Trond Myklebust200baa22006-12-05 00:35:40 -0500652 if (err < 0)
653 result = err;
654 }
Chuck Lever7e381172010-02-01 14:17:41 -0500655 if (result > 0)
656 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657out:
658 return result;
659
660out_swapfile:
661 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
662 goto out;
663}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400664EXPORT_SYMBOL_GPL(nfs_file_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400666ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
667 struct file *filp, loff_t *ppos,
668 size_t count, unsigned int flags)
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700669{
670 struct dentry *dentry = filp->f_path.dentry;
671 struct inode *inode = dentry->d_inode;
Chuck Lever7e381172010-02-01 14:17:41 -0500672 unsigned long written = 0;
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700673 ssize_t ret;
674
675 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
676 dentry->d_parent->d_name.name, dentry->d_name.name,
677 (unsigned long) count, (unsigned long long) *ppos);
678
679 /*
680 * The combination of splice and an O_APPEND destination is disallowed.
681 */
682
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700683 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
Chuck Lever7e381172010-02-01 14:17:41 -0500684 if (ret > 0)
685 written = ret;
686
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700687 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400688 int err = vfs_fsync(filp, 0);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700689 if (err < 0)
690 ret = err;
691 }
Chuck Lever7e381172010-02-01 14:17:41 -0500692 if (ret > 0)
693 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700694 return ret;
695}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400696EXPORT_SYMBOL_GPL(nfs_file_splice_write);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700697
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400698static int
699do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
701 struct inode *inode = filp->f_mapping->host;
702 int status = 0;
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000703 unsigned int saved_type = fl->fl_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Trond Myklebust039c4d72005-10-18 14:20:16 -0700705 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400706 posix_test_lock(filp, fl);
707 if (fl->fl_type != F_UNLCK) {
708 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700709 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 }
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000711 fl->fl_type = saved_type;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700712
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400713 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
Trond Myklebust039c4d72005-10-18 14:20:16 -0700714 goto out_noconflict;
715
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400716 if (is_local)
Trond Myklebust039c4d72005-10-18 14:20:16 -0700717 goto out_noconflict;
718
719 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
720out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700722out_noconflict:
723 fl->fl_type = F_UNLCK;
724 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
727static int do_vfs_lock(struct file *file, struct file_lock *fl)
728{
729 int res = 0;
730 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
731 case FL_POSIX:
732 res = posix_lock_file_wait(file, fl);
733 break;
734 case FL_FLOCK:
735 res = flock_lock_file_wait(file, fl);
736 break;
737 default:
738 BUG();
739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 return res;
741}
742
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400743static int
744do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
746 struct inode *inode = filp->f_mapping->host;
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400747 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 int status;
749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 /*
751 * Flush all pending writes before doing anything
752 * with locks..
753 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500754 nfs_sync_mapping(filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400756 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
757 if (!IS_ERR(l_ctx)) {
758 status = nfs_iocounter_wait(&l_ctx->io_count);
759 nfs_put_lock_context(l_ctx);
760 if (status < 0)
761 return status;
762 }
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 /* NOTE: special case
765 * If we're signalled while cleaning up locks on process exit, we
766 * still need to complete the unlock.
767 */
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400768 /*
769 * Use local locking if mounted with "-onolock" or with appropriate
770 * "-olocal_lock="
771 */
772 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
774 else
775 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 return status;
777}
778
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400779static int
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700780is_time_granular(struct timespec *ts) {
781 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
782}
783
784static int
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400785do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
787 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 int status;
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /*
791 * Flush all pending writes before doing anything
792 * with locks..
793 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500794 status = nfs_sync_mapping(filp->f_mapping);
795 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 goto out;
797
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400798 /*
799 * Use local locking if mounted with "-onolock" or with appropriate
800 * "-olocal_lock="
801 */
802 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400804 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 if (status < 0)
807 goto out;
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700808
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 /*
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700810 * Revalidate the cache if the server has time stamps granular
811 * enough to detect subsecond changes. Otherwise, clear the
812 * cache to prevent missing any changes.
813 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 * This makes locking act as a cache coherency point.
815 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500816 nfs_sync_mapping(filp->f_mapping);
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400817 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700818 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
819 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
820 else
821 nfs_zap_caches(inode);
822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 return status;
825}
826
827/*
828 * Lock a (portion of) a file
829 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400830int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400832 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400833 int ret = -ENOLCK;
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400834 int is_local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Chuck Lever6da24bc2008-06-11 17:55:58 -0400836 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
837 filp->f_path.dentry->d_parent->d_name.name,
838 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 fl->fl_type, fl->fl_flags,
840 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400841
Chuck Lever91d5b472006-03-20 13:44:14 -0500842 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700845 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400846 goto out_err;
847
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400848 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
849 is_local = 1;
850
Trond Myklebust21162712008-05-20 19:34:39 -0400851 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
852 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
853 if (ret < 0)
854 goto out_err;
855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 if (IS_GETLK(cmd))
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400858 ret = do_getlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400859 else if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400860 ret = do_unlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400861 else
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400862 ret = do_setlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400863out_err:
864 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400866EXPORT_SYMBOL_GPL(nfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868/*
869 * Lock a (portion of) a file
870 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400871int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872{
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400873 struct inode *inode = filp->f_mapping->host;
874 int is_local = 0;
875
Chuck Lever6da24bc2008-06-11 17:55:58 -0400876 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
877 filp->f_path.dentry->d_parent->d_name.name,
878 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 fl->fl_type, fl->fl_flags);
880
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 if (!(fl->fl_flags & FL_FLOCK))
882 return -ENOLCK;
883
Jeff Laytonad0fcd42012-07-23 15:46:23 -0400884 /*
885 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
886 * any standard. In principle we might be able to support LOCK_MAND
887 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
888 * NFS code is not set up for it.
889 */
890 if (fl->fl_type & LOCK_MAND)
891 return -EINVAL;
892
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400893 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
894 is_local = 1;
895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 /* We're simulating flock() locks using posix locks on the server */
897 fl->fl_owner = (fl_owner_t)filp;
898 fl->fl_start = 0;
899 fl->fl_end = OFFSET_MAX;
900
901 if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400902 return do_unlk(filp, cmd, fl, is_local);
903 return do_setlk(filp, cmd, fl, is_local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400905EXPORT_SYMBOL_GPL(nfs_flock);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400906
Chuck Lever6da24bc2008-06-11 17:55:58 -0400907/*
908 * There is no protocol support for leases, so we have no way to implement
909 * them correctly in the face of opens by other clients.
910 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400911int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
J. Bruce Fields370f6592007-06-08 15:23:34 -0400912{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400913 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
914 file->f_path.dentry->d_parent->d_name.name,
915 file->f_path.dentry->d_name.name, arg);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400916 return -EINVAL;
917}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400918EXPORT_SYMBOL_GPL(nfs_setlease);
Jeff Layton1788ea62011-11-04 13:31:21 -0400919
Jeff Layton04869582011-11-04 13:31:22 -0400920const struct file_operations nfs_file_operations = {
921 .llseek = nfs_file_llseek,
922 .read = do_sync_read,
923 .write = do_sync_write,
924 .aio_read = nfs_file_read,
925 .aio_write = nfs_file_write,
926 .mmap = nfs_file_mmap,
927 .open = nfs_file_open,
928 .flush = nfs_file_flush,
929 .release = nfs_file_release,
930 .fsync = nfs_file_fsync,
931 .lock = nfs_lock,
932 .flock = nfs_flock,
933 .splice_read = nfs_file_splice_read,
934 .splice_write = nfs_file_splice_write,
935 .check_flags = nfs_check_flags,
936 .setlease = nfs_setlease,
937};
Bryan Schumakerddda8e02012-07-30 16:05:23 -0400938EXPORT_SYMBOL_GPL(nfs_file_operations);