blob: ec7e27d00bc6878a1b9ba6f8f2c4ba24c4cd9b28 [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);
51static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
52 unsigned long nr_segs, loff_t pos);
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -070053static int nfs_file_flush(struct file *, fl_owner_t id);
Chuck Lever54917782008-05-27 16:29:07 -040054static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055static int nfs_check_flags(int flags);
56static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
57static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
J. Bruce Fields370f6592007-06-08 15:23:34 -040058static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Trond Myklebust94387fb2007-07-22 17:09:05 -040060static struct vm_operations_struct nfs_file_vm_ops;
61
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080062const struct file_operations nfs_file_operations = {
Trond Myklebust980802e2005-06-13 11:14:01 -040063 .llseek = nfs_file_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 .read = do_sync_read,
65 .write = do_sync_write,
Badari Pulavarty027445c2006-09-30 23:28:46 -070066 .aio_read = nfs_file_read,
67 .aio_write = nfs_file_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 .mmap = nfs_file_mmap,
69 .open = nfs_file_open,
70 .flush = nfs_file_flush,
71 .release = nfs_file_release,
Chuck Lever54917782008-05-27 16:29:07 -040072 .fsync = nfs_file_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 .lock = nfs_lock,
74 .flock = nfs_flock,
Jens Axboef0930ff2007-06-01 11:51:43 +020075 .splice_read = nfs_file_splice_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 .check_flags = nfs_check_flags,
J. Bruce Fields370f6592007-06-08 15:23:34 -040077 .setlease = nfs_setlease,
Linus Torvalds1da177e2005-04-16 15:20:36 -070078};
79
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080080const struct inode_operations nfs_file_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 .permission = nfs_permission,
82 .getattr = nfs_getattr,
83 .setattr = nfs_setattr,
84};
85
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000086#ifdef CONFIG_NFS_V3
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080087const struct inode_operations nfs3_file_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000088 .permission = nfs_permission,
89 .getattr = nfs_getattr,
90 .setattr = nfs_setattr,
91 .listxattr = nfs3_listxattr,
92 .getxattr = nfs3_getxattr,
93 .setxattr = nfs3_setxattr,
94 .removexattr = nfs3_removexattr,
95};
96#endif /* CONFIG_NFS_v3 */
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098/* Hack for future NFS swap support */
99#ifndef IS_SWAPFILE
100# define IS_SWAPFILE(inode) (0)
101#endif
102
103static int nfs_check_flags(int flags)
104{
105 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
106 return -EINVAL;
107
108 return 0;
109}
110
111/*
112 * Open file
113 */
114static int
115nfs_file_open(struct inode *inode, struct file *filp)
116{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 int res;
118
Chuck Lever6da24bc2008-06-11 17:55:58 -0400119 dprintk("NFS: open file(%s/%s)\n",
Chuck Levercc0dd2d2008-06-11 17:55:42 -0400120 filp->f_path.dentry->d_parent->d_name.name,
121 filp->f_path.dentry->d_name.name);
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 res = nfs_check_flags(filp->f_flags);
124 if (res)
125 return res;
126
Chuck Lever91d5b472006-03-20 13:44:14 -0500127 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Trond Myklebust46cb6502008-06-11 16:32:46 -0400128 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 return res;
130}
131
132static int
133nfs_file_release(struct inode *inode, struct file *filp)
134{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400135 struct dentry *dentry = filp->f_path.dentry;
136
137 dprintk("NFS: release(%s/%s)\n",
138 dentry->d_parent->d_name.name,
139 dentry->d_name.name);
140
Chuck Lever91d5b472006-03-20 13:44:14 -0500141 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Trond Myklebust46cb6502008-06-11 16:32:46 -0400142 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
Trond Myklebust980802e2005-06-13 11:14:01 -0400145/**
146 * nfs_revalidate_size - Revalidate the file size
147 * @inode - pointer to inode struct
148 * @file - pointer to struct file
149 *
150 * Revalidates the file length. This is basically a wrapper around
151 * nfs_revalidate_inode() that takes into account the fact that we may
152 * have cached writes (in which case we don't care about the server's
153 * idea of what the file length is), or O_DIRECT (in which case we
154 * shouldn't trust the cache).
155 */
156static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
157{
158 struct nfs_server *server = NFS_SERVER(inode);
159 struct nfs_inode *nfsi = NFS_I(inode);
160
161 if (server->flags & NFS_MOUNT_NOAC)
162 goto force_reval;
163 if (filp->f_flags & O_DIRECT)
164 goto force_reval;
165 if (nfsi->npages != 0)
166 return 0;
Chuck Lever55296802005-08-18 11:24:09 -0700167 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
Trond Myklebustfe51bee2005-06-22 17:16:30 +0000168 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400169force_reval:
170 return __nfs_revalidate_inode(server, inode);
171}
172
173static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
174{
Andi Kleen9465efc2008-06-27 11:05:24 +0200175 loff_t loff;
Trond Myklebuste89e8962008-07-15 18:34:16 -0400176
Chuck Lever6da24bc2008-06-11 17:55:58 -0400177 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
Chuck Leverb84e06c2008-06-11 17:55:34 -0400178 filp->f_path.dentry->d_parent->d_name.name,
179 filp->f_path.dentry->d_name.name,
180 offset, origin);
181
Trond Myklebust980802e2005-06-13 11:14:01 -0400182 /* origin == SEEK_END => we must revalidate the cached file length */
Josef 'Jeff' Sipekaec5e172006-09-16 21:09:32 -0400183 if (origin == SEEK_END) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400184 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400185
Trond Myklebust980802e2005-06-13 11:14:01 -0400186 int retval = nfs_revalidate_file_size(inode, filp);
187 if (retval < 0)
188 return (loff_t)retval;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400189
190 spin_lock(&inode->i_lock);
191 loff = generic_file_llseek_unlocked(filp, offset, origin);
192 spin_unlock(&inode->i_lock);
193 } else
194 loff = generic_file_llseek_unlocked(filp, offset, origin);
Andi Kleen9465efc2008-06-27 11:05:24 +0200195 return loff;
Trond Myklebust980802e2005-06-13 11:14:01 -0400196}
197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198/*
Chuck Lever54917782008-05-27 16:29:07 -0400199 * Helper for nfs_file_flush() and nfs_file_fsync()
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400200 *
201 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
202 * disk, but it retrieves and clears ctx->error after synching, despite
203 * the two being set at the same time in nfs_context_set_write_error().
204 * This is because the former is used to notify the _next_ call to
205 * nfs_file_write() that a write error occured, and hence cause it to
206 * fall back to doing a synchronous write.
207 */
208static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
209{
210 int have_error, status;
211 int ret = 0;
212
213 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
214 status = nfs_wb_all(inode);
215 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
216 if (have_error)
217 ret = xchg(&ctx->error, 0);
218 if (!ret)
219 ret = status;
220 return ret;
221}
222
223/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 */
226static int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700227nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400229 struct nfs_open_context *ctx = nfs_file_open_context(file);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400230 struct dentry *dentry = file->f_path.dentry;
231 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Chuck Lever6da24bc2008-06-11 17:55:58 -0400233 dprintk("NFS: flush(%s/%s)\n",
234 dentry->d_parent->d_name.name,
235 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237 if ((file->f_mode & FMODE_WRITE) == 0)
238 return 0;
Chuck Lever91d5b472006-03-20 13:44:14 -0500239 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400240
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400241 /* Flush writes to the server and return any errors */
242 return nfs_do_fsync(ctx, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
245static ssize_t
Badari Pulavarty027445c2006-09-30 23:28:46 -0700246nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
247 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800249 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 struct inode * inode = dentry->d_inode;
251 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700252 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700255 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Chuck Lever6da24bc2008-06-11 17:55:58 -0400257 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 dentry->d_parent->d_name.name, dentry->d_name.name,
259 (unsigned long) count, (unsigned long) pos);
260
Trond Myklebust44b11872006-05-25 01:40:59 -0400261 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
Chuck Lever91d5b472006-03-20 13:44:14 -0500262 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (!result)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700264 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 return result;
266}
267
268static ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200269nfs_file_splice_read(struct file *filp, loff_t *ppos,
270 struct pipe_inode_info *pipe, size_t count,
271 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800273 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 struct inode *inode = dentry->d_inode;
275 ssize_t res;
276
Chuck Lever6da24bc2008-06-11 17:55:58 -0400277 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 dentry->d_parent->d_name.name, dentry->d_name.name,
279 (unsigned long) count, (unsigned long long) *ppos);
280
Trond Myklebust44b11872006-05-25 01:40:59 -0400281 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 if (!res)
Jens Axboef0930ff2007-06-01 11:51:43 +0200283 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return res;
285}
286
287static int
288nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
289{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800290 struct dentry *dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 struct inode *inode = dentry->d_inode;
292 int status;
293
Chuck Lever6da24bc2008-06-11 17:55:58 -0400294 dprintk("NFS: mmap(%s/%s)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 dentry->d_parent->d_name.name, dentry->d_name.name);
296
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400297 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
298 * so we call that before revalidating the mapping
299 */
300 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400301 if (!status) {
302 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400303 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400304 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 return status;
306}
307
308/*
309 * Flush any dirty pages for this process, and check for write errors.
310 * The return status from this call provides a reliable indication of
311 * whether any write errors occurred for this process.
312 */
313static int
Chuck Lever54917782008-05-27 16:29:07 -0400314nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400316 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Chuck Lever6da24bc2008-06-11 17:55:58 -0400319 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
Chuck Lever54917782008-05-27 16:29:07 -0400320 dentry->d_parent->d_name.name, dentry->d_name.name,
321 datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Chuck Lever91d5b472006-03-20 13:44:14 -0500323 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400324 return nfs_do_fsync(ctx, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
327/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700328 * This does the "real" work of the write. We must allocate and lock the
329 * page to be sent back to the generic routine, which then copies the
330 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 *
332 * If the writer ends up delaying the write, the writer needs to
333 * increment the page use counts until he is done with the page.
334 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700335static int nfs_write_begin(struct file *file, struct address_space *mapping,
336 loff_t pos, unsigned len, unsigned flags,
337 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700339 int ret;
340 pgoff_t index;
341 struct page *page;
342 index = pos >> PAGE_CACHE_SHIFT;
343
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400344 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
345 file->f_path.dentry->d_parent->d_name.name,
346 file->f_path.dentry->d_name.name,
347 mapping->host->i_ino, len, (long long) pos);
348
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400349 /*
350 * Prevent starvation issues if someone is doing a consistency
351 * sync-to-disk
352 */
353 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
354 nfs_wait_bit_killable, TASK_KILLABLE);
355 if (ret)
356 return ret;
357
Nick Piggin54566b22009-01-04 12:00:53 -0800358 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700359 if (!page)
360 return -ENOMEM;
361 *pagep = page;
362
363 ret = nfs_flush_incompatible(file, page);
364 if (ret) {
365 unlock_page(page);
366 page_cache_release(page);
367 }
368 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
370
Nick Piggin4899f9c2007-10-16 01:25:16 -0700371static int nfs_write_end(struct file *file, struct address_space *mapping,
372 loff_t pos, unsigned len, unsigned copied,
373 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700375 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
376 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400378 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
379 file->f_path.dentry->d_parent->d_name.name,
380 file->f_path.dentry->d_name.name,
381 mapping->host->i_ino, len, (long long) pos);
382
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400383 /*
384 * Zero any uninitialised parts of the page, and then mark the page
385 * as up to date if it turns out that we're extending the file.
386 */
387 if (!PageUptodate(page)) {
388 unsigned pglen = nfs_page_length(page);
389 unsigned end = offset + len;
390
391 if (pglen == 0) {
392 zero_user_segments(page, 0, offset,
393 end, PAGE_CACHE_SIZE);
394 SetPageUptodate(page);
395 } else if (end >= pglen) {
396 zero_user_segment(page, end, PAGE_CACHE_SIZE);
397 if (offset == 0)
398 SetPageUptodate(page);
399 } else
400 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
401 }
402
Nick Piggin4899f9c2007-10-16 01:25:16 -0700403 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700404
405 unlock_page(page);
406 page_cache_release(page);
407
Chuck Lever3d509e52007-12-20 14:55:04 -0500408 if (status < 0)
409 return status;
410 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
412
David Howells6b9b3512009-04-03 16:42:41 +0100413/*
414 * Partially or wholly invalidate a page
415 * - Release the private state associated with a page if undergoing complete
416 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100417 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100418 * - Caller holds page lock
419 */
NeilBrown2ff28e22006-03-26 01:37:18 -0800420static void nfs_invalidate_page(struct page *page, unsigned long offset)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500421{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400422 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
423
Trond Myklebust1c759502006-10-09 16:18:38 -0400424 if (offset != 0)
425 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400426 /* Cancel any unstarted writes on this page */
Trond Myklebust1b3b4a12007-08-28 10:29:36 -0400427 nfs_wb_page_cancel(page->mapping->host, page);
David Howells545db452009-04-03 16:42:44 +0100428
429 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500430}
431
David Howells6b9b3512009-04-03 16:42:41 +0100432/*
433 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100434 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100435 * - Caller holds page lock
436 * - Return true (may release page) or false (may not)
437 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500438static int nfs_release_page(struct page *page, gfp_t gfp)
439{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400440 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
441
Trond Myklebuste3db7692007-01-10 23:15:39 -0800442 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100443 if (PagePrivate(page))
444 return 0;
445 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800446}
447
David Howells6b9b3512009-04-03 16:42:41 +0100448/*
449 * Attempt to clear the private state associated with a page when an error
450 * occurs that requires the cached contents of an inode to be written back or
451 * destroyed
David Howells545db452009-04-03 16:42:44 +0100452 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100453 * - Caller holds page lock
454 * - Return 0 if successful, -error otherwise
455 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800456static int nfs_launder_page(struct page *page)
457{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400458 struct inode *inode = page->mapping->host;
David Howells545db452009-04-03 16:42:44 +0100459 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400460
461 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
462 inode->i_ino, (long long)page_offset(page));
463
David Howells545db452009-04-03 16:42:44 +0100464 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400465 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500466}
467
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700468const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 .readpage = nfs_readpage,
470 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400471 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 .writepage = nfs_writepage,
473 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700474 .write_begin = nfs_write_begin,
475 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500476 .invalidatepage = nfs_invalidate_page,
477 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 .direct_IO = nfs_direct_IO,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800479 .launder_page = nfs_launder_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480};
481
David Howells6b9b3512009-04-03 16:42:41 +0100482/*
483 * Notification that a PTE pointing to an NFS page is about to be made
484 * writable, implying that someone is about to modify the page through a
485 * shared-writable mapping
486 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700487static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400488{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700489 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400490 struct file *filp = vma->vm_file;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400491 struct dentry *dentry = filp->f_path.dentry;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400492 unsigned pagelen;
493 int ret = -EINVAL;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700494 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400495
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400496 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
497 dentry->d_parent->d_name.name, dentry->d_name.name,
498 filp->f_mapping->host->i_ino,
499 (long long)page_offset(page));
500
David Howells545db452009-04-03 16:42:44 +0100501 /* make sure the cache has finished storing the page */
502 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
503
Trond Myklebust94387fb2007-07-22 17:09:05 -0400504 lock_page(page);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700505 mapping = page->mapping;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400506 if (mapping != dentry->d_inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500507 goto out_unlock;
508
509 ret = 0;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400510 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500511 if (pagelen == 0)
512 goto out_unlock;
513
514 ret = nfs_flush_incompatible(filp, page);
515 if (ret != 0)
516 goto out_unlock;
517
518 ret = nfs_updatepage(filp, page, 0, pagelen);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500519out_unlock:
Trond Myklebust7fdf5232009-04-24 17:32:22 -0400520 if (!ret)
521 return VM_FAULT_LOCKED;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400522 unlock_page(page);
Trond Myklebust7fdf5232009-04-24 17:32:22 -0400523 return VM_FAULT_SIGBUS;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400524}
525
526static struct vm_operations_struct nfs_file_vm_ops = {
527 .fault = filemap_fault,
528 .page_mkwrite = nfs_vm_page_mkwrite,
529};
530
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400531static int nfs_need_sync_write(struct file *filp, struct inode *inode)
532{
533 struct nfs_open_context *ctx;
534
535 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
536 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400537 ctx = nfs_file_open_context(filp);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400538 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
539 return 1;
540 return 0;
541}
542
Badari Pulavarty027445c2006-09-30 23:28:46 -0700543static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
544 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800546 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 struct inode * inode = dentry->d_inode;
548 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700549 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700552 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Chuck Lever6da24bc2008-06-11 17:55:58 -0400554 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 dentry->d_parent->d_name.name, dentry->d_name.name,
Chuck Lever6da24bc2008-06-11 17:55:58 -0400556 (unsigned long) count, (long long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 result = -EBUSY;
559 if (IS_SWAPFILE(inode))
560 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000561 /*
562 * O_APPEND implies that we must revalidate the file length.
563 */
564 if (iocb->ki_filp->f_flags & O_APPEND) {
565 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
566 if (result)
567 goto out;
Trond Myklebustfe51bee2005-06-22 17:16:30 +0000568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 result = count;
571 if (!count)
572 goto out;
573
Chuck Lever91d5b472006-03-20 13:44:14 -0500574 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
Badari Pulavarty027445c2006-09-30 23:28:46 -0700575 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
Trond Myklebust200baa22006-12-05 00:35:40 -0500576 /* Return error values for O_SYNC and IS_SYNC() */
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400577 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400578 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
Trond Myklebust200baa22006-12-05 00:35:40 -0500579 if (err < 0)
580 result = err;
581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582out:
583 return result;
584
585out_swapfile:
586 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
587 goto out;
588}
589
590static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
591{
592 struct inode *inode = filp->f_mapping->host;
593 int status = 0;
594
595 lock_kernel();
Trond Myklebust039c4d72005-10-18 14:20:16 -0700596 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400597 posix_test_lock(filp, fl);
598 if (fl->fl_type != F_UNLCK) {
599 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700600 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 }
Trond Myklebust039c4d72005-10-18 14:20:16 -0700602
603 if (nfs_have_delegation(inode, FMODE_READ))
604 goto out_noconflict;
605
606 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
607 goto out_noconflict;
608
609 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
610out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 unlock_kernel();
612 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700613out_noconflict:
614 fl->fl_type = F_UNLCK;
615 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616}
617
618static int do_vfs_lock(struct file *file, struct file_lock *fl)
619{
620 int res = 0;
621 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
622 case FL_POSIX:
623 res = posix_lock_file_wait(file, fl);
624 break;
625 case FL_FLOCK:
626 res = flock_lock_file_wait(file, fl);
627 break;
628 default:
629 BUG();
630 }
631 if (res < 0)
Neil Brown46bae1a2007-01-30 14:36:01 -0800632 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
633 " - error %d!\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700634 __func__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return res;
636}
637
638static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
639{
640 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 int status;
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 /*
644 * Flush all pending writes before doing anything
645 * with locks..
646 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500647 nfs_sync_mapping(filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649 /* NOTE: special case
650 * If we're signalled while cleaning up locks on process exit, we
651 * still need to complete the unlock.
652 */
653 lock_kernel();
654 /* Use local locking if mounted with "-onolock" */
655 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
656 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
657 else
658 status = do_vfs_lock(filp, fl);
659 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 return status;
661}
662
663static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
664{
665 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 int status;
667
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 /*
669 * Flush all pending writes before doing anything
670 * with locks..
671 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500672 status = nfs_sync_mapping(filp->f_mapping);
673 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 goto out;
675
676 lock_kernel();
677 /* Use local locking if mounted with "-onolock" */
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400678 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400680 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 status = do_vfs_lock(filp, fl);
682 unlock_kernel();
683 if (status < 0)
684 goto out;
685 /*
686 * Make sure we clear the cache whenever we try to get the lock.
687 * This makes locking act as a cache coherency point.
688 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500689 nfs_sync_mapping(filp->f_mapping);
Trond Myklebustb5418382008-06-10 18:31:02 -0400690 if (!nfs_have_delegation(inode, FMODE_READ))
691 nfs_zap_caches(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return status;
694}
695
696/*
697 * Lock a (portion of) a file
698 */
699static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
700{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400701 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400702 int ret = -ENOLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Chuck Lever6da24bc2008-06-11 17:55:58 -0400704 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
705 filp->f_path.dentry->d_parent->d_name.name,
706 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 fl->fl_type, fl->fl_flags,
708 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400709
Chuck Lever91d5b472006-03-20 13:44:14 -0500710 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
712 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700713 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400714 goto out_err;
715
716 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
717 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
718 if (ret < 0)
719 goto out_err;
720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722 if (IS_GETLK(cmd))
Trond Myklebust21162712008-05-20 19:34:39 -0400723 ret = do_getlk(filp, cmd, fl);
724 else if (fl->fl_type == F_UNLCK)
725 ret = do_unlk(filp, cmd, fl);
726 else
727 ret = do_setlk(filp, cmd, fl);
728out_err:
729 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
732/*
733 * Lock a (portion of) a file
734 */
735static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
736{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400737 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
738 filp->f_path.dentry->d_parent->d_name.name,
739 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 fl->fl_type, fl->fl_flags);
741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 if (!(fl->fl_flags & FL_FLOCK))
743 return -ENOLCK;
744
745 /* We're simulating flock() locks using posix locks on the server */
746 fl->fl_owner = (fl_owner_t)filp;
747 fl->fl_start = 0;
748 fl->fl_end = OFFSET_MAX;
749
750 if (fl->fl_type == F_UNLCK)
751 return do_unlk(filp, cmd, fl);
752 return do_setlk(filp, cmd, fl);
753}
J. Bruce Fields370f6592007-06-08 15:23:34 -0400754
Chuck Lever6da24bc2008-06-11 17:55:58 -0400755/*
756 * There is no protocol support for leases, so we have no way to implement
757 * them correctly in the face of opens by other clients.
758 */
J. Bruce Fields370f6592007-06-08 15:23:34 -0400759static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
760{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400761 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
762 file->f_path.dentry->d_parent->d_name.name,
763 file->f_path.dentry->d_name.name, arg);
764
J. Bruce Fields370f6592007-06-08 15:23:34 -0400765 return -EINVAL;
766}