blob: 0055b813ec2c7004e9647ad781c3706066aeac3e [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
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/errno.h>
22#include <linux/fcntl.h>
23#include <linux/stat.h>
24#include <linux/nfs_fs.h>
25#include <linux/nfs_mount.h>
26#include <linux/mm.h>
27#include <linux/slab.h>
28#include <linux/pagemap.h>
29#include <linux/smp_lock.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040030#include <linux/aio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/uaccess.h>
33#include <asm/system.h>
34
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
42static int nfs_file_open(struct inode *, struct file *);
43static int nfs_file_release(struct inode *, struct file *);
Trond Myklebust980802e2005-06-13 11:14:01 -040044static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045static int nfs_file_mmap(struct file *, struct vm_area_struct *);
Jens Axboef0930ff2007-06-01 11:51:43 +020046static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
47 struct pipe_inode_info *pipe,
48 size_t count, unsigned int flags);
Badari Pulavarty027445c2006-09-30 23:28:46 -070049static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
50 unsigned long nr_segs, loff_t pos);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -070051static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
52 struct file *filp, loff_t *ppos,
53 size_t count, unsigned int flags);
Badari Pulavarty027445c2006-09-30 23:28:46 -070054static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
55 unsigned long nr_segs, loff_t pos);
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -070056static int nfs_file_flush(struct file *, fl_owner_t id);
Chuck Lever54917782008-05-27 16:29:07 -040057static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static int nfs_check_flags(int flags);
59static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
60static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
J. Bruce Fields370f6592007-06-08 15:23:34 -040061static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Trond Myklebust94387fb2007-07-22 17:09:05 -040063static struct vm_operations_struct nfs_file_vm_ops;
64
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080065const struct file_operations nfs_file_operations = {
Trond Myklebust980802e2005-06-13 11:14:01 -040066 .llseek = nfs_file_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 .read = do_sync_read,
68 .write = do_sync_write,
Badari Pulavarty027445c2006-09-30 23:28:46 -070069 .aio_read = nfs_file_read,
70 .aio_write = nfs_file_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 .mmap = nfs_file_mmap,
72 .open = nfs_file_open,
73 .flush = nfs_file_flush,
74 .release = nfs_file_release,
Chuck Lever54917782008-05-27 16:29:07 -040075 .fsync = nfs_file_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 .lock = nfs_lock,
77 .flock = nfs_flock,
Jens Axboef0930ff2007-06-01 11:51:43 +020078 .splice_read = nfs_file_splice_read,
Suresh Jayaramanbf40d342009-06-17 18:02:09 -070079 .splice_write = nfs_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 .check_flags = nfs_check_flags,
J. Bruce Fields370f6592007-06-08 15:23:34 -040081 .setlease = nfs_setlease,
Linus Torvalds1da177e2005-04-16 15:20:36 -070082};
83
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080084const struct inode_operations nfs_file_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 .permission = nfs_permission,
86 .getattr = nfs_getattr,
87 .setattr = nfs_setattr,
88};
89
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000090#ifdef CONFIG_NFS_V3
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080091const struct inode_operations nfs3_file_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000092 .permission = nfs_permission,
93 .getattr = nfs_getattr,
94 .setattr = nfs_setattr,
95 .listxattr = nfs3_listxattr,
96 .getxattr = nfs3_getxattr,
97 .setxattr = nfs3_setxattr,
98 .removexattr = nfs3_removexattr,
99};
100#endif /* CONFIG_NFS_v3 */
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/* Hack for future NFS swap support */
103#ifndef IS_SWAPFILE
104# define IS_SWAPFILE(inode) (0)
105#endif
106
107static int nfs_check_flags(int flags)
108{
109 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
110 return -EINVAL;
111
112 return 0;
113}
114
115/*
116 * Open file
117 */
118static int
119nfs_file_open(struct inode *inode, struct file *filp)
120{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 int res;
122
Chuck Lever6da24bc2008-06-11 17:55:58 -0400123 dprintk("NFS: open file(%s/%s)\n",
Chuck Levercc0dd2d2008-06-11 17:55:42 -0400124 filp->f_path.dentry->d_parent->d_name.name,
125 filp->f_path.dentry->d_name.name);
126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 res = nfs_check_flags(filp->f_flags);
128 if (res)
129 return res;
130
Chuck Lever91d5b472006-03-20 13:44:14 -0500131 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Trond Myklebust46cb6502008-06-11 16:32:46 -0400132 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 return res;
134}
135
136static int
137nfs_file_release(struct inode *inode, struct file *filp)
138{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400139 struct dentry *dentry = filp->f_path.dentry;
140
141 dprintk("NFS: release(%s/%s)\n",
142 dentry->d_parent->d_name.name,
143 dentry->d_name.name);
144
Chuck Lever91d5b472006-03-20 13:44:14 -0500145 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Trond Myklebust46cb6502008-06-11 16:32:46 -0400146 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Trond Myklebust980802e2005-06-13 11:14:01 -0400149/**
150 * nfs_revalidate_size - Revalidate the file size
151 * @inode - pointer to inode struct
152 * @file - pointer to struct file
153 *
154 * Revalidates the file length. This is basically a wrapper around
155 * nfs_revalidate_inode() that takes into account the fact that we may
156 * have cached writes (in which case we don't care about the server's
157 * idea of what the file length is), or O_DIRECT (in which case we
158 * shouldn't trust the cache).
159 */
160static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
161{
162 struct nfs_server *server = NFS_SERVER(inode);
163 struct nfs_inode *nfsi = NFS_I(inode);
164
165 if (server->flags & NFS_MOUNT_NOAC)
166 goto force_reval;
167 if (filp->f_flags & O_DIRECT)
168 goto force_reval;
169 if (nfsi->npages != 0)
170 return 0;
Chuck Lever55296802005-08-18 11:24:09 -0700171 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000172 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400173force_reval:
174 return __nfs_revalidate_inode(server, inode);
175}
176
177static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
178{
Andi Kleen9465efc2008-06-27 11:05:24 +0200179 loff_t loff;
Trond Myklebuste89e8962008-07-15 18:34:16 -0400180
Chuck Lever6da24bc2008-06-11 17:55:58 -0400181 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
Chuck Leverb84e06c2008-06-11 17:55:34 -0400182 filp->f_path.dentry->d_parent->d_name.name,
183 filp->f_path.dentry->d_name.name,
184 offset, origin);
185
Trond Myklebust980802e2005-06-13 11:14:01 -0400186 /* origin == SEEK_END => we must revalidate the cached file length */
Josef 'Jeff' Sipekaec5e172006-09-16 21:09:32 -0400187 if (origin == SEEK_END) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400188 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400189
Trond Myklebust980802e2005-06-13 11:14:01 -0400190 int retval = nfs_revalidate_file_size(inode, filp);
191 if (retval < 0)
192 return (loff_t)retval;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400193
194 spin_lock(&inode->i_lock);
195 loff = generic_file_llseek_unlocked(filp, offset, origin);
196 spin_unlock(&inode->i_lock);
197 } else
198 loff = generic_file_llseek_unlocked(filp, offset, origin);
Andi Kleen9465efc2008-06-27 11:05:24 +0200199 return loff;
Trond Myklebust980802e2005-06-13 11:14:01 -0400200}
201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202/*
Chuck Lever54917782008-05-27 16:29:07 -0400203 * Helper for nfs_file_flush() and nfs_file_fsync()
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400204 *
205 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
206 * disk, but it retrieves and clears ctx->error after synching, despite
207 * the two being set at the same time in nfs_context_set_write_error().
208 * This is because the former is used to notify the _next_ call to
209 * nfs_file_write() that a write error occured, and hence cause it to
210 * fall back to doing a synchronous write.
211 */
212static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
213{
214 int have_error, status;
215 int ret = 0;
216
217 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
218 status = nfs_wb_all(inode);
219 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
220 if (have_error)
221 ret = xchg(&ctx->error, 0);
222 if (!ret)
223 ret = status;
224 return ret;
225}
226
227/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
230static int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700231nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400233 struct nfs_open_context *ctx = nfs_file_open_context(file);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400234 struct dentry *dentry = file->f_path.dentry;
235 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Chuck Lever6da24bc2008-06-11 17:55:58 -0400237 dprintk("NFS: flush(%s/%s)\n",
238 dentry->d_parent->d_name.name,
239 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
241 if ((file->f_mode & FMODE_WRITE) == 0)
242 return 0;
Chuck Lever91d5b472006-03-20 13:44:14 -0500243 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400244
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400245 /* Flush writes to the server and return any errors */
246 return nfs_do_fsync(ctx, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
249static ssize_t
Badari Pulavarty027445c2006-09-30 23:28:46 -0700250nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
251 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800253 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 struct inode * inode = dentry->d_inode;
255 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700256 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700259 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Chuck Lever6da24bc2008-06-11 17:55:58 -0400261 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 dentry->d_parent->d_name.name, dentry->d_name.name,
263 (unsigned long) count, (unsigned long) pos);
264
Trond Myklebust44b11872006-05-25 01:40:59 -0400265 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
Chuck Lever91d5b472006-03-20 13:44:14 -0500266 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 if (!result)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700268 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return result;
270}
271
272static ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200273nfs_file_splice_read(struct file *filp, loff_t *ppos,
274 struct pipe_inode_info *pipe, size_t count,
275 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800277 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 struct inode *inode = dentry->d_inode;
279 ssize_t res;
280
Chuck Lever6da24bc2008-06-11 17:55:58 -0400281 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 dentry->d_parent->d_name.name, dentry->d_name.name,
283 (unsigned long) count, (unsigned long long) *ppos);
284
Trond Myklebust44b11872006-05-25 01:40:59 -0400285 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 if (!res)
Jens Axboef0930ff2007-06-01 11:51:43 +0200287 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return res;
289}
290
291static int
292nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
293{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800294 struct dentry *dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 struct inode *inode = dentry->d_inode;
296 int status;
297
Chuck Lever6da24bc2008-06-11 17:55:58 -0400298 dprintk("NFS: mmap(%s/%s)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 dentry->d_parent->d_name.name, dentry->d_name.name);
300
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400301 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
302 * so we call that before revalidating the mapping
303 */
304 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400305 if (!status) {
306 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400307 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 return status;
310}
311
312/*
313 * Flush any dirty pages for this process, and check for write errors.
314 * The return status from this call provides a reliable indication of
315 * whether any write errors occurred for this process.
316 */
317static int
Chuck Lever54917782008-05-27 16:29:07 -0400318nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400320 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Chuck Lever6da24bc2008-06-11 17:55:58 -0400323 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
Chuck Lever54917782008-05-27 16:29:07 -0400324 dentry->d_parent->d_name.name, dentry->d_name.name,
325 datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Chuck Lever91d5b472006-03-20 13:44:14 -0500327 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400328 return nfs_do_fsync(ctx, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329}
330
331/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700332 * This does the "real" work of the write. We must allocate and lock the
333 * page to be sent back to the generic routine, which then copies the
334 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 *
336 * If the writer ends up delaying the write, the writer needs to
337 * increment the page use counts until he is done with the page.
338 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700339static int nfs_write_begin(struct file *file, struct address_space *mapping,
340 loff_t pos, unsigned len, unsigned flags,
341 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700343 int ret;
344 pgoff_t index;
345 struct page *page;
346 index = pos >> PAGE_CACHE_SHIFT;
347
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400348 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
349 file->f_path.dentry->d_parent->d_name.name,
350 file->f_path.dentry->d_name.name,
351 mapping->host->i_ino, len, (long long) pos);
352
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400353 /*
354 * Prevent starvation issues if someone is doing a consistency
355 * sync-to-disk
356 */
357 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
358 nfs_wait_bit_killable, TASK_KILLABLE);
359 if (ret)
360 return ret;
361
Nick Piggin54566b22009-01-04 12:00:53 -0800362 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700363 if (!page)
364 return -ENOMEM;
365 *pagep = page;
366
367 ret = nfs_flush_incompatible(file, page);
368 if (ret) {
369 unlock_page(page);
370 page_cache_release(page);
371 }
372 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
Nick Piggin4899f9c2007-10-16 01:25:16 -0700375static int nfs_write_end(struct file *file, struct address_space *mapping,
376 loff_t pos, unsigned len, unsigned copied,
377 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700379 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
380 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400382 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
383 file->f_path.dentry->d_parent->d_name.name,
384 file->f_path.dentry->d_name.name,
385 mapping->host->i_ino, len, (long long) pos);
386
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400387 /*
388 * Zero any uninitialised parts of the page, and then mark the page
389 * as up to date if it turns out that we're extending the file.
390 */
391 if (!PageUptodate(page)) {
392 unsigned pglen = nfs_page_length(page);
393 unsigned end = offset + len;
394
395 if (pglen == 0) {
396 zero_user_segments(page, 0, offset,
397 end, PAGE_CACHE_SIZE);
398 SetPageUptodate(page);
399 } else if (end >= pglen) {
400 zero_user_segment(page, end, PAGE_CACHE_SIZE);
401 if (offset == 0)
402 SetPageUptodate(page);
403 } else
404 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
405 }
406
Nick Piggin4899f9c2007-10-16 01:25:16 -0700407 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700408
409 unlock_page(page);
410 page_cache_release(page);
411
Chuck Lever3d509e52007-12-20 14:55:04 -0500412 if (status < 0)
413 return status;
414 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
David Howells6b9b3512009-04-03 16:42:41 +0100417/*
418 * Partially or wholly invalidate a page
419 * - Release the private state associated with a page if undergoing complete
420 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100421 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100422 * - Caller holds page lock
423 */
NeilBrown2ff28e22006-03-26 01:37:18 -0800424static void nfs_invalidate_page(struct page *page, unsigned long offset)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500425{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400426 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
427
Trond Myklebust1c759502006-10-09 16:18:38 -0400428 if (offset != 0)
429 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400430 /* Cancel any unstarted writes on this page */
Trond Myklebust1b3b4a12007-08-28 10:29:36 -0400431 nfs_wb_page_cancel(page->mapping->host, page);
David Howells545db452009-04-03 16:42:44 +0100432
433 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500434}
435
David Howells6b9b3512009-04-03 16:42:41 +0100436/*
437 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100438 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100439 * - Caller holds page lock
440 * - Return true (may release page) or false (may not)
441 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500442static int nfs_release_page(struct page *page, gfp_t gfp)
443{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400444 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
445
Trond Myklebuste3db7692007-01-10 23:15:39 -0800446 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100447 if (PagePrivate(page))
448 return 0;
449 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800450}
451
David Howells6b9b3512009-04-03 16:42:41 +0100452/*
453 * Attempt to clear the private state associated with a page when an error
454 * occurs that requires the cached contents of an inode to be written back or
455 * destroyed
David Howells545db452009-04-03 16:42:44 +0100456 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100457 * - Caller holds page lock
458 * - Return 0 if successful, -error otherwise
459 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800460static int nfs_launder_page(struct page *page)
461{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400462 struct inode *inode = page->mapping->host;
David Howells545db452009-04-03 16:42:44 +0100463 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400464
465 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
466 inode->i_ino, (long long)page_offset(page));
467
David Howells545db452009-04-03 16:42:44 +0100468 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400469 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500470}
471
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700472const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 .readpage = nfs_readpage,
474 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400475 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 .writepage = nfs_writepage,
477 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700478 .write_begin = nfs_write_begin,
479 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500480 .invalidatepage = nfs_invalidate_page,
481 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 .direct_IO = nfs_direct_IO,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800483 .launder_page = nfs_launder_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484};
485
David Howells6b9b3512009-04-03 16:42:41 +0100486/*
487 * Notification that a PTE pointing to an NFS page is about to be made
488 * writable, implying that someone is about to modify the page through a
489 * shared-writable mapping
490 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700491static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400492{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700493 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400494 struct file *filp = vma->vm_file;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400495 struct dentry *dentry = filp->f_path.dentry;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400496 unsigned pagelen;
497 int ret = -EINVAL;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700498 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400499
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400500 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
501 dentry->d_parent->d_name.name, dentry->d_name.name,
502 filp->f_mapping->host->i_ino,
503 (long long)page_offset(page));
504
David Howells545db452009-04-03 16:42:44 +0100505 /* make sure the cache has finished storing the page */
506 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
507
Trond Myklebust94387fb2007-07-22 17:09:05 -0400508 lock_page(page);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700509 mapping = page->mapping;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400510 if (mapping != dentry->d_inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500511 goto out_unlock;
512
513 ret = 0;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400514 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500515 if (pagelen == 0)
516 goto out_unlock;
517
518 ret = nfs_flush_incompatible(filp, page);
519 if (ret != 0)
520 goto out_unlock;
521
522 ret = nfs_updatepage(filp, page, 0, pagelen);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500523out_unlock:
Trond Myklebust7fdf5232009-04-24 17:32:22 -0400524 if (!ret)
525 return VM_FAULT_LOCKED;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400526 unlock_page(page);
Trond Myklebust7fdf5232009-04-24 17:32:22 -0400527 return VM_FAULT_SIGBUS;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400528}
529
530static struct vm_operations_struct nfs_file_vm_ops = {
531 .fault = filemap_fault,
532 .page_mkwrite = nfs_vm_page_mkwrite,
533};
534
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400535static int nfs_need_sync_write(struct file *filp, struct inode *inode)
536{
537 struct nfs_open_context *ctx;
538
539 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
540 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400541 ctx = nfs_file_open_context(filp);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400542 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
543 return 1;
544 return 0;
545}
546
Badari Pulavarty027445c2006-09-30 23:28:46 -0700547static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
548 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800550 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 struct inode * inode = dentry->d_inode;
552 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700553 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700556 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Chuck Lever6da24bc2008-06-11 17:55:58 -0400558 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 dentry->d_parent->d_name.name, dentry->d_name.name,
Chuck Lever6da24bc2008-06-11 17:55:58 -0400560 (unsigned long) count, (long long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562 result = -EBUSY;
563 if (IS_SWAPFILE(inode))
564 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000565 /*
566 * O_APPEND implies that we must revalidate the file length.
567 */
568 if (iocb->ki_filp->f_flags & O_APPEND) {
569 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
570 if (result)
571 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
574 result = count;
575 if (!count)
576 goto out;
577
Chuck Lever91d5b472006-03-20 13:44:14 -0500578 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
Badari Pulavarty027445c2006-09-30 23:28:46 -0700579 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
Trond Myklebust200baa22006-12-05 00:35:40 -0500580 /* Return error values for O_SYNC and IS_SYNC() */
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400581 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400582 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
Trond Myklebust200baa22006-12-05 00:35:40 -0500583 if (err < 0)
584 result = err;
585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586out:
587 return result;
588
589out_swapfile:
590 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
591 goto out;
592}
593
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700594static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
595 struct file *filp, loff_t *ppos,
596 size_t count, unsigned int flags)
597{
598 struct dentry *dentry = filp->f_path.dentry;
599 struct inode *inode = dentry->d_inode;
600 ssize_t ret;
601
602 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
603 dentry->d_parent->d_name.name, dentry->d_name.name,
604 (unsigned long) count, (unsigned long long) *ppos);
605
606 /*
607 * The combination of splice and an O_APPEND destination is disallowed.
608 */
609
610 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
611
612 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
613 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
614 int err = nfs_do_fsync(nfs_file_open_context(filp), inode);
615 if (err < 0)
616 ret = err;
617 }
618 return ret;
619}
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
622{
623 struct inode *inode = filp->f_mapping->host;
624 int status = 0;
625
Trond Myklebust039c4d72005-10-18 14:20:16 -0700626 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400627 posix_test_lock(filp, fl);
628 if (fl->fl_type != F_UNLCK) {
629 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700630 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
Trond Myklebust039c4d72005-10-18 14:20:16 -0700632
633 if (nfs_have_delegation(inode, FMODE_READ))
634 goto out_noconflict;
635
636 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
637 goto out_noconflict;
638
639 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
640out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700642out_noconflict:
643 fl->fl_type = F_UNLCK;
644 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647static int do_vfs_lock(struct file *file, struct file_lock *fl)
648{
649 int res = 0;
650 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
651 case FL_POSIX:
652 res = posix_lock_file_wait(file, fl);
653 break;
654 case FL_FLOCK:
655 res = flock_lock_file_wait(file, fl);
656 break;
657 default:
658 BUG();
659 }
660 if (res < 0)
Neil Brown46bae1a2007-01-30 14:36:01 -0800661 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
662 " - error %d!\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700663 __func__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 return res;
665}
666
667static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
668{
669 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 int status;
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 /*
673 * Flush all pending writes before doing anything
674 * with locks..
675 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500676 nfs_sync_mapping(filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678 /* NOTE: special case
679 * If we're signalled while cleaning up locks on process exit, we
680 * still need to complete the unlock.
681 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 /* Use local locking if mounted with "-onolock" */
683 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
684 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
685 else
686 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 return status;
688}
689
690static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
691{
692 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 int status;
694
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 /*
696 * Flush all pending writes before doing anything
697 * with locks..
698 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500699 status = nfs_sync_mapping(filp->f_mapping);
700 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 goto out;
702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 /* Use local locking if mounted with "-onolock" */
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400704 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400706 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (status < 0)
709 goto out;
710 /*
711 * Make sure we clear the cache whenever we try to get the lock.
712 * This makes locking act as a cache coherency point.
713 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500714 nfs_sync_mapping(filp->f_mapping);
Trond Myklebustb5418382008-06-10 18:31:02 -0400715 if (!nfs_have_delegation(inode, FMODE_READ))
716 nfs_zap_caches(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 return status;
719}
720
721/*
722 * Lock a (portion of) a file
723 */
724static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
725{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400726 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400727 int ret = -ENOLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Chuck Lever6da24bc2008-06-11 17:55:58 -0400729 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
730 filp->f_path.dentry->d_parent->d_name.name,
731 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 fl->fl_type, fl->fl_flags,
733 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400734
Chuck Lever91d5b472006-03-20 13:44:14 -0500735 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
737 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700738 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400739 goto out_err;
740
741 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
742 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
743 if (ret < 0)
744 goto out_err;
745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
747 if (IS_GETLK(cmd))
Trond Myklebust21162712008-05-20 19:34:39 -0400748 ret = do_getlk(filp, cmd, fl);
749 else if (fl->fl_type == F_UNLCK)
750 ret = do_unlk(filp, cmd, fl);
751 else
752 ret = do_setlk(filp, cmd, fl);
753out_err:
754 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755}
756
757/*
758 * Lock a (portion of) a file
759 */
760static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
761{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400762 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
763 filp->f_path.dentry->d_parent->d_name.name,
764 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 fl->fl_type, fl->fl_flags);
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 if (!(fl->fl_flags & FL_FLOCK))
768 return -ENOLCK;
769
770 /* We're simulating flock() locks using posix locks on the server */
771 fl->fl_owner = (fl_owner_t)filp;
772 fl->fl_start = 0;
773 fl->fl_end = OFFSET_MAX;
774
775 if (fl->fl_type == F_UNLCK)
776 return do_unlk(filp, cmd, fl);
777 return do_setlk(filp, cmd, fl);
778}
J. Bruce Fields370f6592007-06-08 15:23:34 -0400779
Chuck Lever6da24bc2008-06-11 17:55:58 -0400780/*
781 * There is no protocol support for leases, so we have no way to implement
782 * them correctly in the face of opens by other clients.
783 */
J. Bruce Fields370f6592007-06-08 15:23:34 -0400784static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
785{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400786 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
787 file->f_path.dentry->d_parent->d_name.name,
788 file->f_path.dentry->d_name.name, arg);
789
J. Bruce Fields370f6592007-06-08 15:23:34 -0400790 return -EINVAL;
791}