blob: 05062329b678a148c464990b0f9d4f6a68ab617f [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>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040029#include <linux/aio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31#include <asm/uaccess.h>
32#include <asm/system.h>
33
34#include "delegation.h"
Trond Myklebust94387fb2007-07-22 17:09:05 -040035#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050036#include "iostat.h"
David Howells545db452009-04-03 16:42:44 +010037#include "fscache.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#define NFSDBG_FACILITY NFSDBG_FILE
40
41static int nfs_file_open(struct inode *, struct file *);
42static int nfs_file_release(struct inode *, struct file *);
Trond Myklebust980802e2005-06-13 11:14:01 -040043static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static int nfs_file_mmap(struct file *, struct vm_area_struct *);
Jens Axboef0930ff2007-06-01 11:51:43 +020045static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
46 struct pipe_inode_info *pipe,
47 size_t count, unsigned int flags);
Badari Pulavarty027445c2006-09-30 23:28:46 -070048static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
49 unsigned long nr_segs, loff_t pos);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -070050static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
51 struct file *filp, loff_t *ppos,
52 size_t count, unsigned int flags);
Badari Pulavarty027445c2006-09-30 23:28:46 -070053static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
54 unsigned long nr_segs, loff_t pos);
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -070055static int nfs_file_flush(struct file *, fl_owner_t id);
Chuck Lever54917782008-05-27 16:29:07 -040056static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057static int nfs_check_flags(int flags);
58static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
59static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
J. Bruce Fields370f6592007-06-08 15:23:34 -040060static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Trond Myklebust94387fb2007-07-22 17:09:05 -040062static struct vm_operations_struct nfs_file_vm_ops;
63
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080064const struct file_operations nfs_file_operations = {
Trond Myklebust980802e2005-06-13 11:14:01 -040065 .llseek = nfs_file_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 .read = do_sync_read,
67 .write = do_sync_write,
Badari Pulavarty027445c2006-09-30 23:28:46 -070068 .aio_read = nfs_file_read,
69 .aio_write = nfs_file_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 .mmap = nfs_file_mmap,
71 .open = nfs_file_open,
72 .flush = nfs_file_flush,
73 .release = nfs_file_release,
Chuck Lever54917782008-05-27 16:29:07 -040074 .fsync = nfs_file_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 .lock = nfs_lock,
76 .flock = nfs_flock,
Jens Axboef0930ff2007-06-01 11:51:43 +020077 .splice_read = nfs_file_splice_read,
Suresh Jayaramanbf40d342009-06-17 18:02:09 -070078 .splice_write = nfs_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 .check_flags = nfs_check_flags,
J. Bruce Fields370f6592007-06-08 15:23:34 -040080 .setlease = nfs_setlease,
Linus Torvalds1da177e2005-04-16 15:20:36 -070081};
82
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080083const struct inode_operations nfs_file_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 .permission = nfs_permission,
85 .getattr = nfs_getattr,
86 .setattr = nfs_setattr,
87};
88
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000089#ifdef CONFIG_NFS_V3
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080090const struct inode_operations nfs3_file_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000091 .permission = nfs_permission,
92 .getattr = nfs_getattr,
93 .setattr = nfs_setattr,
94 .listxattr = nfs3_listxattr,
95 .getxattr = nfs3_getxattr,
96 .setxattr = nfs3_setxattr,
97 .removexattr = nfs3_removexattr,
98};
99#endif /* CONFIG_NFS_v3 */
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101/* Hack for future NFS swap support */
102#ifndef IS_SWAPFILE
103# define IS_SWAPFILE(inode) (0)
104#endif
105
106static int nfs_check_flags(int flags)
107{
108 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
109 return -EINVAL;
110
111 return 0;
112}
113
114/*
115 * Open file
116 */
117static int
118nfs_file_open(struct inode *inode, struct file *filp)
119{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 int res;
121
Chuck Lever6da24bc2008-06-11 17:55:58 -0400122 dprintk("NFS: open file(%s/%s)\n",
Chuck Levercc0dd2d2008-06-11 17:55:42 -0400123 filp->f_path.dentry->d_parent->d_name.name,
124 filp->f_path.dentry->d_name.name);
125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 res = nfs_check_flags(filp->f_flags);
127 if (res)
128 return res;
129
Chuck Lever91d5b472006-03-20 13:44:14 -0500130 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Trond Myklebust46cb6502008-06-11 16:32:46 -0400131 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 return res;
133}
134
135static int
136nfs_file_release(struct inode *inode, struct file *filp)
137{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400138 struct dentry *dentry = filp->f_path.dentry;
139
140 dprintk("NFS: release(%s/%s)\n",
141 dentry->d_parent->d_name.name,
142 dentry->d_name.name);
143
Chuck Lever91d5b472006-03-20 13:44:14 -0500144 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Trond Myklebust46cb6502008-06-11 16:32:46 -0400145 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
Trond Myklebust980802e2005-06-13 11:14:01 -0400148/**
149 * nfs_revalidate_size - Revalidate the file size
150 * @inode - pointer to inode struct
151 * @file - pointer to struct file
152 *
153 * Revalidates the file length. This is basically a wrapper around
154 * nfs_revalidate_inode() that takes into account the fact that we may
155 * have cached writes (in which case we don't care about the server's
156 * idea of what the file length is), or O_DIRECT (in which case we
157 * shouldn't trust the cache).
158 */
159static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
160{
161 struct nfs_server *server = NFS_SERVER(inode);
162 struct nfs_inode *nfsi = NFS_I(inode);
163
164 if (server->flags & NFS_MOUNT_NOAC)
165 goto force_reval;
166 if (filp->f_flags & O_DIRECT)
167 goto force_reval;
168 if (nfsi->npages != 0)
169 return 0;
Chuck Lever55296802005-08-18 11:24:09 -0700170 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000171 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400172force_reval:
173 return __nfs_revalidate_inode(server, inode);
174}
175
176static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
177{
Andi Kleen9465efc2008-06-27 11:05:24 +0200178 loff_t loff;
Trond Myklebuste89e8962008-07-15 18:34:16 -0400179
Chuck Lever6da24bc2008-06-11 17:55:58 -0400180 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
Chuck Leverb84e06c2008-06-11 17:55:34 -0400181 filp->f_path.dentry->d_parent->d_name.name,
182 filp->f_path.dentry->d_name.name,
183 offset, origin);
184
Trond Myklebust980802e2005-06-13 11:14:01 -0400185 /* origin == SEEK_END => we must revalidate the cached file length */
Josef 'Jeff' Sipekaec5e172006-09-16 21:09:32 -0400186 if (origin == SEEK_END) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400187 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400188
Trond Myklebust980802e2005-06-13 11:14:01 -0400189 int retval = nfs_revalidate_file_size(inode, filp);
190 if (retval < 0)
191 return (loff_t)retval;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400192
193 spin_lock(&inode->i_lock);
194 loff = generic_file_llseek_unlocked(filp, offset, origin);
195 spin_unlock(&inode->i_lock);
196 } else
197 loff = generic_file_llseek_unlocked(filp, offset, origin);
Andi Kleen9465efc2008-06-27 11:05:24 +0200198 return loff;
Trond Myklebust980802e2005-06-13 11:14:01 -0400199}
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201/*
Chuck Lever54917782008-05-27 16:29:07 -0400202 * Helper for nfs_file_flush() and nfs_file_fsync()
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400203 *
204 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
205 * disk, but it retrieves and clears ctx->error after synching, despite
206 * the two being set at the same time in nfs_context_set_write_error().
207 * This is because the former is used to notify the _next_ call to
208 * nfs_file_write() that a write error occured, and hence cause it to
209 * fall back to doing a synchronous write.
210 */
211static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
212{
213 int have_error, status;
214 int ret = 0;
215
216 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
217 status = nfs_wb_all(inode);
218 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
219 if (have_error)
220 ret = xchg(&ctx->error, 0);
221 if (!ret)
222 ret = status;
223 return ret;
224}
225
226/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 */
229static int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700230nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400232 struct nfs_open_context *ctx = nfs_file_open_context(file);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400233 struct dentry *dentry = file->f_path.dentry;
234 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Chuck Lever6da24bc2008-06-11 17:55:58 -0400236 dprintk("NFS: flush(%s/%s)\n",
237 dentry->d_parent->d_name.name,
238 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240 if ((file->f_mode & FMODE_WRITE) == 0)
241 return 0;
Chuck Lever91d5b472006-03-20 13:44:14 -0500242 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400243
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400244 /* Flush writes to the server and return any errors */
245 return nfs_do_fsync(ctx, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
248static ssize_t
Badari Pulavarty027445c2006-09-30 23:28:46 -0700249nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
250 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800252 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 struct inode * inode = dentry->d_inode;
254 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700255 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700258 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Chuck Lever6da24bc2008-06-11 17:55:58 -0400260 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 dentry->d_parent->d_name.name, dentry->d_name.name,
262 (unsigned long) count, (unsigned long) pos);
263
Trond Myklebust44b11872006-05-25 01:40:59 -0400264 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
Chuck Lever91d5b472006-03-20 13:44:14 -0500265 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 if (!result)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700267 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 return result;
269}
270
271static ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200272nfs_file_splice_read(struct file *filp, loff_t *ppos,
273 struct pipe_inode_info *pipe, size_t count,
274 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800276 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 struct inode *inode = dentry->d_inode;
278 ssize_t res;
279
Chuck Lever6da24bc2008-06-11 17:55:58 -0400280 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 dentry->d_parent->d_name.name, dentry->d_name.name,
282 (unsigned long) count, (unsigned long long) *ppos);
283
Trond Myklebust44b11872006-05-25 01:40:59 -0400284 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 if (!res)
Jens Axboef0930ff2007-06-01 11:51:43 +0200286 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return res;
288}
289
290static int
291nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
292{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800293 struct dentry *dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 struct inode *inode = dentry->d_inode;
295 int status;
296
Chuck Lever6da24bc2008-06-11 17:55:58 -0400297 dprintk("NFS: mmap(%s/%s)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 dentry->d_parent->d_name.name, dentry->d_name.name);
299
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400300 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
301 * so we call that before revalidating the mapping
302 */
303 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400304 if (!status) {
305 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400306 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 return status;
309}
310
311/*
312 * Flush any dirty pages for this process, and check for write errors.
313 * The return status from this call provides a reliable indication of
314 * whether any write errors occurred for this process.
315 */
316static int
Chuck Lever54917782008-05-27 16:29:07 -0400317nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400319 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Chuck Lever6da24bc2008-06-11 17:55:58 -0400322 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
Chuck Lever54917782008-05-27 16:29:07 -0400323 dentry->d_parent->d_name.name, dentry->d_name.name,
324 datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
Chuck Lever91d5b472006-03-20 13:44:14 -0500326 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400327 return nfs_do_fsync(ctx, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328}
329
330/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700331 * This does the "real" work of the write. We must allocate and lock the
332 * page to be sent back to the generic routine, which then copies the
333 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 *
335 * If the writer ends up delaying the write, the writer needs to
336 * increment the page use counts until he is done with the page.
337 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700338static int nfs_write_begin(struct file *file, struct address_space *mapping,
339 loff_t pos, unsigned len, unsigned flags,
340 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700342 int ret;
343 pgoff_t index;
344 struct page *page;
345 index = pos >> PAGE_CACHE_SHIFT;
346
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400347 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
348 file->f_path.dentry->d_parent->d_name.name,
349 file->f_path.dentry->d_name.name,
350 mapping->host->i_ino, len, (long long) pos);
351
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400352 /*
353 * Prevent starvation issues if someone is doing a consistency
354 * sync-to-disk
355 */
356 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
357 nfs_wait_bit_killable, TASK_KILLABLE);
358 if (ret)
359 return ret;
360
Nick Piggin54566b22009-01-04 12:00:53 -0800361 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700362 if (!page)
363 return -ENOMEM;
364 *pagep = page;
365
366 ret = nfs_flush_incompatible(file, page);
367 if (ret) {
368 unlock_page(page);
369 page_cache_release(page);
370 }
371 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
Nick Piggin4899f9c2007-10-16 01:25:16 -0700374static int nfs_write_end(struct file *file, struct address_space *mapping,
375 loff_t pos, unsigned len, unsigned copied,
376 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700378 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
379 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400381 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
382 file->f_path.dentry->d_parent->d_name.name,
383 file->f_path.dentry->d_name.name,
384 mapping->host->i_ino, len, (long long) pos);
385
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400386 /*
387 * Zero any uninitialised parts of the page, and then mark the page
388 * as up to date if it turns out that we're extending the file.
389 */
390 if (!PageUptodate(page)) {
391 unsigned pglen = nfs_page_length(page);
392 unsigned end = offset + len;
393
394 if (pglen == 0) {
395 zero_user_segments(page, 0, offset,
396 end, PAGE_CACHE_SIZE);
397 SetPageUptodate(page);
398 } else if (end >= pglen) {
399 zero_user_segment(page, end, PAGE_CACHE_SIZE);
400 if (offset == 0)
401 SetPageUptodate(page);
402 } else
403 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
404 }
405
Nick Piggin4899f9c2007-10-16 01:25:16 -0700406 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700407
408 unlock_page(page);
409 page_cache_release(page);
410
Chuck Lever3d509e52007-12-20 14:55:04 -0500411 if (status < 0)
412 return status;
413 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
David Howells6b9b3512009-04-03 16:42:41 +0100416/*
417 * Partially or wholly invalidate a page
418 * - Release the private state associated with a page if undergoing complete
419 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100420 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100421 * - Caller holds page lock
422 */
NeilBrown2ff28e22006-03-26 01:37:18 -0800423static void nfs_invalidate_page(struct page *page, unsigned long offset)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500424{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400425 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
426
Trond Myklebust1c759502006-10-09 16:18:38 -0400427 if (offset != 0)
428 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400429 /* Cancel any unstarted writes on this page */
Trond Myklebust1b3b4a12007-08-28 10:29:36 -0400430 nfs_wb_page_cancel(page->mapping->host, page);
David Howells545db452009-04-03 16:42:44 +0100431
432 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500433}
434
David Howells6b9b3512009-04-03 16:42:41 +0100435/*
436 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100437 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100438 * - Caller holds page lock
439 * - Return true (may release page) or false (may not)
440 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500441static int nfs_release_page(struct page *page, gfp_t gfp)
442{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400443 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
444
Trond Myklebuste3db7692007-01-10 23:15:39 -0800445 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100446 if (PagePrivate(page))
447 return 0;
448 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800449}
450
David Howells6b9b3512009-04-03 16:42:41 +0100451/*
452 * Attempt to clear the private state associated with a page when an error
453 * occurs that requires the cached contents of an inode to be written back or
454 * destroyed
David Howells545db452009-04-03 16:42:44 +0100455 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100456 * - Caller holds page lock
457 * - Return 0 if successful, -error otherwise
458 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800459static int nfs_launder_page(struct page *page)
460{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400461 struct inode *inode = page->mapping->host;
David Howells545db452009-04-03 16:42:44 +0100462 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400463
464 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
465 inode->i_ino, (long long)page_offset(page));
466
David Howells545db452009-04-03 16:42:44 +0100467 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400468 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500469}
470
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700471const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 .readpage = nfs_readpage,
473 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400474 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 .writepage = nfs_writepage,
476 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700477 .write_begin = nfs_write_begin,
478 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500479 .invalidatepage = nfs_invalidate_page,
480 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 .direct_IO = nfs_direct_IO,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800482 .launder_page = nfs_launder_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483};
484
David Howells6b9b3512009-04-03 16:42:41 +0100485/*
486 * Notification that a PTE pointing to an NFS page is about to be made
487 * writable, implying that someone is about to modify the page through a
488 * shared-writable mapping
489 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700490static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400491{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700492 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400493 struct file *filp = vma->vm_file;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400494 struct dentry *dentry = filp->f_path.dentry;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400495 unsigned pagelen;
496 int ret = -EINVAL;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700497 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400498
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400499 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
500 dentry->d_parent->d_name.name, dentry->d_name.name,
501 filp->f_mapping->host->i_ino,
502 (long long)page_offset(page));
503
David Howells545db452009-04-03 16:42:44 +0100504 /* make sure the cache has finished storing the page */
505 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
506
Trond Myklebust94387fb2007-07-22 17:09:05 -0400507 lock_page(page);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700508 mapping = page->mapping;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400509 if (mapping != dentry->d_inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500510 goto out_unlock;
511
512 ret = 0;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400513 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500514 if (pagelen == 0)
515 goto out_unlock;
516
517 ret = nfs_flush_incompatible(filp, page);
518 if (ret != 0)
519 goto out_unlock;
520
521 ret = nfs_updatepage(filp, page, 0, pagelen);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500522out_unlock:
Trond Myklebust7fdf5232009-04-24 17:32:22 -0400523 if (!ret)
524 return VM_FAULT_LOCKED;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400525 unlock_page(page);
Trond Myklebust7fdf5232009-04-24 17:32:22 -0400526 return VM_FAULT_SIGBUS;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400527}
528
529static struct vm_operations_struct nfs_file_vm_ops = {
530 .fault = filemap_fault,
531 .page_mkwrite = nfs_vm_page_mkwrite,
532};
533
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400534static int nfs_need_sync_write(struct file *filp, struct inode *inode)
535{
536 struct nfs_open_context *ctx;
537
538 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
539 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400540 ctx = nfs_file_open_context(filp);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400541 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
542 return 1;
543 return 0;
544}
545
Badari Pulavarty027445c2006-09-30 23:28:46 -0700546static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
547 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800549 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 struct inode * inode = dentry->d_inode;
551 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700552 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700555 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Chuck Lever6da24bc2008-06-11 17:55:58 -0400557 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 dentry->d_parent->d_name.name, dentry->d_name.name,
Chuck Lever6da24bc2008-06-11 17:55:58 -0400559 (unsigned long) count, (long long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 result = -EBUSY;
562 if (IS_SWAPFILE(inode))
563 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000564 /*
565 * O_APPEND implies that we must revalidate the file length.
566 */
567 if (iocb->ki_filp->f_flags & O_APPEND) {
568 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
569 if (result)
570 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 result = count;
574 if (!count)
575 goto out;
576
Chuck Lever91d5b472006-03-20 13:44:14 -0500577 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
Badari Pulavarty027445c2006-09-30 23:28:46 -0700578 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
Trond Myklebust200baa22006-12-05 00:35:40 -0500579 /* Return error values for O_SYNC and IS_SYNC() */
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400580 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400581 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
Trond Myklebust200baa22006-12-05 00:35:40 -0500582 if (err < 0)
583 result = err;
584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585out:
586 return result;
587
588out_swapfile:
589 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
590 goto out;
591}
592
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700593static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
594 struct file *filp, loff_t *ppos,
595 size_t count, unsigned int flags)
596{
597 struct dentry *dentry = filp->f_path.dentry;
598 struct inode *inode = dentry->d_inode;
599 ssize_t ret;
600
601 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
602 dentry->d_parent->d_name.name, dentry->d_name.name,
603 (unsigned long) count, (unsigned long long) *ppos);
604
605 /*
606 * The combination of splice and an O_APPEND destination is disallowed.
607 */
608
609 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
610
611 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
612 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
613 int err = nfs_do_fsync(nfs_file_open_context(filp), inode);
614 if (err < 0)
615 ret = err;
616 }
617 return ret;
618}
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
621{
622 struct inode *inode = filp->f_mapping->host;
623 int status = 0;
624
Trond Myklebust039c4d72005-10-18 14:20:16 -0700625 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400626 posix_test_lock(filp, fl);
627 if (fl->fl_type != F_UNLCK) {
628 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700629 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
Trond Myklebust039c4d72005-10-18 14:20:16 -0700631
632 if (nfs_have_delegation(inode, FMODE_READ))
633 goto out_noconflict;
634
635 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
636 goto out_noconflict;
637
638 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
639out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700641out_noconflict:
642 fl->fl_type = F_UNLCK;
643 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
646static int do_vfs_lock(struct file *file, struct file_lock *fl)
647{
648 int res = 0;
649 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
650 case FL_POSIX:
651 res = posix_lock_file_wait(file, fl);
652 break;
653 case FL_FLOCK:
654 res = flock_lock_file_wait(file, fl);
655 break;
656 default:
657 BUG();
658 }
659 if (res < 0)
Neil Brown46bae1a2007-01-30 14:36:01 -0800660 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
661 " - error %d!\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700662 __func__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 return res;
664}
665
666static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
667{
668 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 int status;
670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 /*
672 * Flush all pending writes before doing anything
673 * with locks..
674 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500675 nfs_sync_mapping(filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 /* NOTE: special case
678 * If we're signalled while cleaning up locks on process exit, we
679 * still need to complete the unlock.
680 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 /* Use local locking if mounted with "-onolock" */
682 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
683 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
684 else
685 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 return status;
687}
688
689static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
690{
691 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 int status;
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 /*
695 * Flush all pending writes before doing anything
696 * with locks..
697 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500698 status = nfs_sync_mapping(filp->f_mapping);
699 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 goto out;
701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 /* Use local locking if mounted with "-onolock" */
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400703 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400705 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 if (status < 0)
708 goto out;
709 /*
710 * Make sure we clear the cache whenever we try to get the lock.
711 * This makes locking act as a cache coherency point.
712 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500713 nfs_sync_mapping(filp->f_mapping);
Trond Myklebustb5418382008-06-10 18:31:02 -0400714 if (!nfs_have_delegation(inode, FMODE_READ))
715 nfs_zap_caches(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 return status;
718}
719
720/*
721 * Lock a (portion of) a file
722 */
723static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
724{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400725 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400726 int ret = -ENOLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Chuck Lever6da24bc2008-06-11 17:55:58 -0400728 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
729 filp->f_path.dentry->d_parent->d_name.name,
730 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 fl->fl_type, fl->fl_flags,
732 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400733
Chuck Lever91d5b472006-03-20 13:44:14 -0500734 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700737 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400738 goto out_err;
739
740 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
741 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
742 if (ret < 0)
743 goto out_err;
744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746 if (IS_GETLK(cmd))
Trond Myklebust21162712008-05-20 19:34:39 -0400747 ret = do_getlk(filp, cmd, fl);
748 else if (fl->fl_type == F_UNLCK)
749 ret = do_unlk(filp, cmd, fl);
750 else
751 ret = do_setlk(filp, cmd, fl);
752out_err:
753 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
755
756/*
757 * Lock a (portion of) a file
758 */
759static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
760{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400761 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
762 filp->f_path.dentry->d_parent->d_name.name,
763 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 fl->fl_type, fl->fl_flags);
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 if (!(fl->fl_flags & FL_FLOCK))
767 return -ENOLCK;
768
769 /* We're simulating flock() locks using posix locks on the server */
770 fl->fl_owner = (fl_owner_t)filp;
771 fl->fl_start = 0;
772 fl->fl_end = OFFSET_MAX;
773
774 if (fl->fl_type == F_UNLCK)
775 return do_unlk(filp, cmd, fl);
776 return do_setlk(filp, cmd, fl);
777}
J. Bruce Fields370f6592007-06-08 15:23:34 -0400778
Chuck Lever6da24bc2008-06-11 17:55:58 -0400779/*
780 * There is no protocol support for leases, so we have no way to implement
781 * them correctly in the face of opens by other clients.
782 */
J. Bruce Fields370f6592007-06-08 15:23:34 -0400783static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
784{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400785 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
786 file->f_path.dentry->d_parent->d_name.name,
787 file->f_path.dentry->d_name.name, arg);
788
J. Bruce Fields370f6592007-06-08 15:23:34 -0400789 return -EINVAL;
790}