blob: 89bf093d342a583d0d3c19738a2baebf046c68e2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/file.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
11 *
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
13 *
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
15 *
16 * nfs regular file handling functions
17 */
18
Bryan Schumakerddda8e02012-07-30 16:05:23 -040019#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/time.h>
21#include <linux/kernel.h>
22#include <linux/errno.h>
23#include <linux/fcntl.h>
24#include <linux/stat.h>
25#include <linux/nfs_fs.h>
26#include <linux/nfs_mount.h>
27#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/pagemap.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090029#include <linux/gfp.h>
Trond Myklebustb608b282010-07-30 15:31:54 -040030#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
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"
Christoph Hellwig612aa982014-09-10 08:23:30 -070038#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Trond Myklebustf4ce1292013-08-19 18:59:33 -040040#include "nfstrace.h"
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#define NFSDBG_FACILITY NFSDBG_FILE
43
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040044static const struct vm_operations_struct nfs_file_vm_ops;
Trond Myklebust94387fb2007-07-22 17:09:05 -040045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046/* Hack for future NFS swap support */
47#ifndef IS_SWAPFILE
48# define IS_SWAPFILE(inode) (0)
49#endif
50
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -040051int nfs_check_flags(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
54 return -EINVAL;
55
56 return 0;
57}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040058EXPORT_SYMBOL_GPL(nfs_check_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
61 * Open file
62 */
63static int
64nfs_file_open(struct inode *inode, struct file *filp)
65{
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 int res;
67
Al Viro6de14722013-09-16 10:53:17 -040068 dprintk("NFS: open file(%pD2)\n", filp);
Chuck Levercc0dd2d2008-06-11 17:55:42 -040069
Chuck Leverc2459dc2010-02-01 14:17:14 -050070 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 res = nfs_check_flags(filp->f_flags);
72 if (res)
73 return res;
74
Trond Myklebust46cb6502008-06-11 16:32:46 -040075 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 return res;
77}
78
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -040079int
Linus Torvalds1da177e2005-04-16 15:20:36 -070080nfs_file_release(struct inode *inode, struct file *filp)
81{
Al Viro6de14722013-09-16 10:53:17 -040082 dprintk("NFS: release(%pD2)\n", filp);
Chuck Lever6da24bc2008-06-11 17:55:58 -040083
Chuck Lever91d5b472006-03-20 13:44:14 -050084 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Anna Schumakeraff8d8d2015-07-13 14:01:33 -040085 nfs_file_clear_open_context(filp);
86 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040088EXPORT_SYMBOL_GPL(nfs_file_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Trond Myklebust980802e2005-06-13 11:14:01 -040090/**
91 * nfs_revalidate_size - Revalidate the file size
92 * @inode - pointer to inode struct
93 * @file - pointer to struct file
94 *
95 * Revalidates the file length. This is basically a wrapper around
96 * nfs_revalidate_inode() that takes into account the fact that we may
97 * have cached writes (in which case we don't care about the server's
98 * idea of what the file length is), or O_DIRECT (in which case we
99 * shouldn't trust the cache).
100 */
101static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
102{
103 struct nfs_server *server = NFS_SERVER(inode);
104 struct nfs_inode *nfsi = NFS_I(inode);
105
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400106 if (nfs_have_delegated_attributes(inode))
107 goto out_noreval;
108
Trond Myklebust980802e2005-06-13 11:14:01 -0400109 if (filp->f_flags & O_DIRECT)
110 goto force_reval;
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400111 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
112 goto force_reval;
113 if (nfs_attribute_timeout(inode))
114 goto force_reval;
115out_noreval:
116 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400117force_reval:
118 return __nfs_revalidate_inode(server, inode);
119}
120
Andrew Morton965c8e52012-12-17 15:59:39 -0800121loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
Trond Myklebust980802e2005-06-13 11:14:01 -0400122{
Al Viro6de14722013-09-16 10:53:17 -0400123 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
124 filp, offset, whence);
Chuck Leverb84e06c2008-06-11 17:55:34 -0400125
Josef Bacik06222e42011-07-18 13:21:38 -0400126 /*
Andrew Morton965c8e52012-12-17 15:59:39 -0800127 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
Josef Bacik06222e42011-07-18 13:21:38 -0400128 * the cached file length
129 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800130 if (whence != SEEK_SET && whence != SEEK_CUR) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400131 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400132
Trond Myklebust980802e2005-06-13 11:14:01 -0400133 int retval = nfs_revalidate_file_size(inode, filp);
134 if (retval < 0)
135 return (loff_t)retval;
Andi Kleen79835a712011-09-15 16:06:52 -0700136 }
Trond Myklebustd5e66342008-09-23 17:28:35 -0400137
Andrew Morton965c8e52012-12-17 15:59:39 -0800138 return generic_file_llseek(filp, offset, whence);
Trond Myklebust980802e2005-06-13 11:14:01 -0400139}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400140EXPORT_SYMBOL_GPL(nfs_file_llseek);
Trond Myklebust980802e2005-06-13 11:14:01 -0400141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142/*
143 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 */
Trond Myklebust5445b1f2015-09-05 19:06:58 -0400145static int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700146nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147{
Al Viro6de14722013-09-16 10:53:17 -0400148 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Al Viro6de14722013-09-16 10:53:17 -0400150 dprintk("NFS: flush(%pD2)\n", file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Chuck Leverc2459dc2010-02-01 14:17:14 -0500152 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 if ((file->f_mode & FMODE_WRITE) == 0)
154 return 0;
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400155
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400156 /* Flush writes to the server and return any errors */
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400157 return vfs_fsync(file, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400160ssize_t
Al Viro3aa2d192014-04-02 20:14:12 -0400161nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
Al Viro6de14722013-09-16 10:53:17 -0400163 struct inode *inode = file_inode(iocb->ki_filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 ssize_t result;
165
Al Viro2ba48ce2015-04-09 13:52:01 -0400166 if (iocb->ki_flags & IOCB_DIRECT)
Martin K. Petersene19a8a02014-10-14 09:00:44 -0600167 return nfs_file_direct_read(iocb, to, iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Al Viro619d30b2014-03-04 21:53:33 -0500169 dprintk("NFS: read(%pD2, %zu@%lu)\n",
Al Viro6de14722013-09-16 10:53:17 -0400170 iocb->ki_filp,
Al Viro3aa2d192014-04-02 20:14:12 -0400171 iov_iter_count(to), (unsigned long) iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Trond Myklebust874f9462015-03-02 23:32:08 -0500173 result = nfs_revalidate_mapping_protected(inode, iocb->ki_filp->f_mapping);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500174 if (!result) {
Al Viro3aa2d192014-04-02 20:14:12 -0400175 result = generic_file_read_iter(iocb, to);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500176 if (result > 0)
177 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 return result;
180}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400181EXPORT_SYMBOL_GPL(nfs_file_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400183ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200184nfs_file_splice_read(struct file *filp, loff_t *ppos,
185 struct pipe_inode_info *pipe, size_t count,
186 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
Al Viro6de14722013-09-16 10:53:17 -0400188 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 ssize_t res;
190
Al Viro6de14722013-09-16 10:53:17 -0400191 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
192 filp, (unsigned long) count, (unsigned long long) *ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Trond Myklebust874f9462015-03-02 23:32:08 -0500194 res = nfs_revalidate_mapping_protected(inode, filp->f_mapping);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500195 if (!res) {
Jens Axboef0930ff2007-06-01 11:51:43 +0200196 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500197 if (res > 0)
198 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
199 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 return res;
201}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400202EXPORT_SYMBOL_GPL(nfs_file_splice_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400204int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
206{
Al Viro6de14722013-09-16 10:53:17 -0400207 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 int status;
209
Al Viro6de14722013-09-16 10:53:17 -0400210 dprintk("NFS: mmap(%pD2)\n", file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400212 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
213 * so we call that before revalidating the mapping
214 */
215 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400216 if (!status) {
217 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400218 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 return status;
221}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400222EXPORT_SYMBOL_GPL(nfs_file_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
224/*
225 * Flush any dirty pages for this process, and check for write errors.
226 * The return status from this call provides a reliable indication of
227 * whether any write errors occurred for this process.
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400228 *
229 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
230 * disk, but it retrieves and clears ctx->error after synching, despite
231 * the two being set at the same time in nfs_context_set_write_error().
232 * This is because the former is used to notify the _next_ call to
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300233 * nfs_file_write() that a write error occurred, and hence cause it to
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400234 * fall back to doing a synchronous write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 */
Christoph Hellwig4ff79bc2016-03-02 17:35:54 +0100236static int
Bryan Schumakera5c58892012-06-20 15:53:42 -0400237nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400239 struct nfs_open_context *ctx = nfs_file_open_context(file);
Al Viro6de14722013-09-16 10:53:17 -0400240 struct inode *inode = file_inode(file);
Trond Myklebust05990d12012-09-11 16:01:22 -0400241 int have_error, do_resend, status;
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400242 int ret = 0;
243
Al Viro6de14722013-09-16 10:53:17 -0400244 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Chuck Lever91d5b472006-03-20 13:44:14 -0500246 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebust05990d12012-09-11 16:01:22 -0400247 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400248 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
249 status = nfs_commit_inode(inode, FLUSH_SYNC);
250 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
Trond Myklebust05990d12012-09-11 16:01:22 -0400251 if (have_error) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400252 ret = xchg(&ctx->error, 0);
Trond Myklebust05990d12012-09-11 16:01:22 -0400253 if (ret)
254 goto out;
255 }
256 if (status < 0) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400257 ret = status;
Trond Myklebust05990d12012-09-11 16:01:22 -0400258 goto out;
259 }
260 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
261 if (do_resend)
262 ret = -EAGAIN;
263out:
Bryan Schumakera5c58892012-06-20 15:53:42 -0400264 return ret;
265}
266
Christoph Hellwig4ff79bc2016-03-02 17:35:54 +0100267int
Bryan Schumakera5c58892012-06-20 15:53:42 -0400268nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
269{
270 int ret;
Al Viro496ad9a2013-01-23 17:07:38 -0500271 struct inode *inode = file_inode(file);
Bryan Schumakera5c58892012-06-20 15:53:42 -0400272
Trond Myklebustf4ce1292013-08-19 18:59:33 -0400273 trace_nfs_fsync_enter(inode);
274
Christoph Hellwig95d9f6c2016-03-02 17:35:55 +0100275 inode_dio_wait(inode);
Trond Myklebust05990d12012-09-11 16:01:22 -0400276 do {
277 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
278 if (ret != 0)
279 break;
Al Viro59551022016-01-22 15:40:57 -0500280 inode_lock(inode);
Trond Myklebust05990d12012-09-11 16:01:22 -0400281 ret = nfs_file_fsync_commit(file, start, end, datasync);
Christoph Hellwig4ff79bc2016-03-02 17:35:54 +0100282 if (!ret)
283 ret = pnfs_sync_inode(inode, !!datasync);
Al Viro59551022016-01-22 15:40:57 -0500284 inode_unlock(inode);
Trond Myklebustdcfc4f22012-09-11 16:19:38 -0400285 /*
286 * If nfs_file_fsync_commit detected a server reboot, then
287 * resend all dirty pages that might have been covered by
288 * the NFS_CONTEXT_RESEND_WRITES flag
289 */
290 start = 0;
291 end = LLONG_MAX;
Trond Myklebust05990d12012-09-11 16:01:22 -0400292 } while (ret == -EAGAIN);
293
Trond Myklebustf4ce1292013-08-19 18:59:33 -0400294 trace_nfs_fsync_exit(inode, ret);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400295 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
Christoph Hellwig4ff79bc2016-03-02 17:35:54 +0100297EXPORT_SYMBOL_GPL(nfs_file_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
299/*
Peter Staubach38c73042009-08-10 08:54:16 -0400300 * Decide whether a read/modify/write cycle may be more efficient
301 * then a modify/write/read cycle when writing to a page in the
302 * page cache.
303 *
304 * The modify/write/read cycle may occur if a page is read before
305 * being completely filled by the writer. In this situation, the
306 * page must be completely written to stable storage on the server
307 * before it can be refilled by reading in the page from the server.
308 * This can lead to expensive, small, FILE_SYNC mode writes being
309 * done.
310 *
311 * It may be more efficient to read the page first if the file is
312 * open for reading in addition to writing, the page is not marked
313 * as Uptodate, it is not dirty or waiting to be committed,
314 * indicating that it was previously allocated and then modified,
315 * that there were valid bytes of data in that range of the file,
316 * and that the new data won't completely replace the old data in
317 * that range of the file.
318 */
319static int nfs_want_read_modify_write(struct file *file, struct page *page,
320 loff_t pos, unsigned len)
321{
322 unsigned int pglen = nfs_page_length(page);
323 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
324 unsigned int end = offset + len;
325
Christoph Hellwig612aa982014-09-10 08:23:30 -0700326 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
327 if (!PageUptodate(page))
328 return 1;
329 return 0;
330 }
331
Peter Staubach38c73042009-08-10 08:54:16 -0400332 if ((file->f_mode & FMODE_READ) && /* open for read? */
333 !PageUptodate(page) && /* Uptodate? */
334 !PagePrivate(page) && /* i/o request already? */
335 pglen && /* valid bytes of file? */
336 (end < pglen || offset)) /* replace all valid bytes? */
337 return 1;
338 return 0;
339}
340
341/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700342 * This does the "real" work of the write. We must allocate and lock the
343 * page to be sent back to the generic routine, which then copies the
344 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 *
346 * If the writer ends up delaying the write, the writer needs to
347 * increment the page use counts until he is done with the page.
348 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700349static int nfs_write_begin(struct file *file, struct address_space *mapping,
350 loff_t pos, unsigned len, unsigned flags,
351 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700353 int ret;
Peter Staubach38c73042009-08-10 08:54:16 -0400354 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700355 struct page *page;
Peter Staubach38c73042009-08-10 08:54:16 -0400356 int once_thru = 0;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700357
Niels de Vos1e8968c2013-12-17 18:20:16 +0100358 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400359 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400360
Peter Staubach38c73042009-08-10 08:54:16 -0400361start:
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400362 /*
363 * Prevent starvation issues if someone is doing a consistency
364 * sync-to-disk
365 */
NeilBrown74316202014-07-07 15:16:04 +1000366 ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
367 nfs_wait_bit_killable, TASK_KILLABLE);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400368 if (ret)
369 return ret;
Trond Myklebustaa5accea2015-02-27 22:54:19 -0500370 /*
371 * Wait for O_DIRECT to complete
372 */
Christoph Hellwig95d9f6c2016-03-02 17:35:55 +0100373 inode_dio_wait(mapping->host);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400374
Nick Piggin54566b22009-01-04 12:00:53 -0800375 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700376 if (!page)
377 return -ENOMEM;
378 *pagep = page;
379
380 ret = nfs_flush_incompatible(file, page);
381 if (ret) {
382 unlock_page(page);
383 page_cache_release(page);
Peter Staubach38c73042009-08-10 08:54:16 -0400384 } else if (!once_thru &&
385 nfs_want_read_modify_write(file, page, pos, len)) {
386 once_thru = 1;
387 ret = nfs_readpage(file, page);
388 page_cache_release(page);
389 if (!ret)
390 goto start;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700391 }
392 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393}
394
Nick Piggin4899f9c2007-10-16 01:25:16 -0700395static int nfs_write_end(struct file *file, struct address_space *mapping,
396 loff_t pos, unsigned len, unsigned copied,
397 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700399 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
Andy Adamsondc248262013-08-14 11:59:16 -0400400 struct nfs_open_context *ctx = nfs_file_open_context(file);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700401 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Niels de Vos1e8968c2013-12-17 18:20:16 +0100403 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400404 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400405
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400406 /*
407 * Zero any uninitialised parts of the page, and then mark the page
408 * as up to date if it turns out that we're extending the file.
409 */
410 if (!PageUptodate(page)) {
411 unsigned pglen = nfs_page_length(page);
412 unsigned end = offset + len;
413
414 if (pglen == 0) {
415 zero_user_segments(page, 0, offset,
416 end, PAGE_CACHE_SIZE);
417 SetPageUptodate(page);
418 } else if (end >= pglen) {
419 zero_user_segment(page, end, PAGE_CACHE_SIZE);
420 if (offset == 0)
421 SetPageUptodate(page);
422 } else
423 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
424 }
425
Nick Piggin4899f9c2007-10-16 01:25:16 -0700426 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700427
428 unlock_page(page);
429 page_cache_release(page);
430
Chuck Lever3d509e52007-12-20 14:55:04 -0500431 if (status < 0)
432 return status;
Andy Adamson2701d082012-05-24 13:13:24 -0400433 NFS_I(mapping->host)->write_io += copied;
Andy Adamsondc248262013-08-14 11:59:16 -0400434
435 if (nfs_ctx_key_to_expire(ctx)) {
436 status = nfs_wb_all(mapping->host);
437 if (status < 0)
438 return status;
439 }
440
Chuck Lever3d509e52007-12-20 14:55:04 -0500441 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443
David Howells6b9b3512009-04-03 16:42:41 +0100444/*
445 * Partially or wholly invalidate a page
446 * - Release the private state associated with a page if undergoing complete
447 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100448 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100449 * - Caller holds page lock
450 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400451static void nfs_invalidate_page(struct page *page, unsigned int offset,
452 unsigned int length)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500453{
Lukas Czernerd47992f2013-05-21 23:17:23 -0400454 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
455 page, offset, length);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400456
Lukas Czernerd47992f2013-05-21 23:17:23 -0400457 if (offset != 0 || length < PAGE_CACHE_SIZE)
Trond Myklebust1c759502006-10-09 16:18:38 -0400458 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400459 /* Cancel any unstarted writes on this page */
Mel Gormand56b4dd2012-07-31 16:45:06 -0700460 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
David Howells545db452009-04-03 16:42:44 +0100461
462 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500463}
464
David Howells6b9b3512009-04-03 16:42:41 +0100465/*
466 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100467 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100468 * - Caller holds page lock
469 * - Return true (may release page) or false (may not)
470 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500471static int nfs_release_page(struct page *page, gfp_t gfp)
472{
Trond Myklebustb608b282010-07-30 15:31:54 -0400473 struct address_space *mapping = page->mapping;
474
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400475 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
476
NeilBrown95905442014-09-24 11:28:32 +1000477 /* Always try to initiate a 'commit' if relevant, but only
Mel Gormand0164ad2015-11-06 16:28:21 -0800478 * wait for it if the caller allows blocking. Even then,
479 * only wait 1 second and only if the 'bdi' is not congested.
NeilBrown95905442014-09-24 11:28:32 +1000480 * Waiting indefinitely can cause deadlocks when the NFS
NeilBrown1aff5252014-09-24 11:28:32 +1000481 * server is on this machine, when a new TCP connection is
482 * needed and in other rare cases. There is no particular
483 * need to wait extensively here. A short wait has the
484 * benefit that someone else can worry about the freezer.
Jeff Layton5cf02d02012-07-23 13:58:51 -0400485 */
NeilBrown95905442014-09-24 11:28:32 +1000486 if (mapping) {
487 struct nfs_server *nfss = NFS_SERVER(mapping->host);
488 nfs_commit_inode(mapping->host, 0);
Mel Gormand0164ad2015-11-06 16:28:21 -0800489 if (gfpflags_allow_blocking(gfp) &&
NeilBrown353db792014-09-24 11:28:32 +1000490 !bdi_write_congested(&nfss->backing_dev_info)) {
NeilBrown95905442014-09-24 11:28:32 +1000491 wait_on_page_bit_killable_timeout(page, PG_private,
492 HZ);
NeilBrown353db792014-09-24 11:28:32 +1000493 if (PagePrivate(page))
494 set_bdi_congested(&nfss->backing_dev_info,
495 BLK_RW_ASYNC);
NeilBrown95905442014-09-24 11:28:32 +1000496 }
Trond Myklebustb608b282010-07-30 15:31:54 -0400497 }
Trond Myklebuste3db7692007-01-10 23:15:39 -0800498 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100499 if (PagePrivate(page))
500 return 0;
501 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800502}
503
Mel Gormanf919b192013-07-03 15:02:06 -0700504static void nfs_check_dirty_writeback(struct page *page,
505 bool *dirty, bool *writeback)
506{
507 struct nfs_inode *nfsi;
508 struct address_space *mapping = page_file_mapping(page);
509
510 if (!mapping || PageSwapCache(page))
511 return;
512
513 /*
514 * Check if an unstable page is currently being committed and
515 * if so, have the VM treat it as if the page is under writeback
516 * so it will not block due to pages that will shortly be freeable.
517 */
518 nfsi = NFS_I(mapping->host);
Trond Myklebustaf7cf052015-09-29 20:34:05 -0400519 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
Mel Gormanf919b192013-07-03 15:02:06 -0700520 *writeback = true;
521 return;
522 }
523
524 /*
525 * If PagePrivate() is set, then the page is not freeable and as the
526 * inode is not being committed, it's not going to be cleaned in the
527 * near future so treat it as dirty
528 */
529 if (PagePrivate(page))
530 *dirty = true;
531}
532
David Howells6b9b3512009-04-03 16:42:41 +0100533/*
534 * Attempt to clear the private state associated with a page when an error
535 * occurs that requires the cached contents of an inode to be written back or
536 * destroyed
David Howells545db452009-04-03 16:42:44 +0100537 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100538 * - Caller holds page lock
539 * - Return 0 if successful, -error otherwise
540 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800541static int nfs_launder_page(struct page *page)
542{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700543 struct inode *inode = page_file_mapping(page)->host;
David Howells545db452009-04-03 16:42:44 +0100544 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400545
546 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
547 inode->i_ino, (long long)page_offset(page));
548
David Howells545db452009-04-03 16:42:44 +0100549 nfs_fscache_wait_on_page_write(nfsi, page);
Peng Taod6c843b2015-12-05 16:20:43 +0800550 return nfs_wb_launder_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500551}
552
Mel Gormana564b8f2012-07-31 16:45:12 -0700553static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
554 sector_t *span)
555{
Jeff Laytondad2b012014-09-10 09:03:55 -0400556 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
557
Mel Gormana564b8f2012-07-31 16:45:12 -0700558 *span = sis->pages;
Jeff Laytondad2b012014-09-10 09:03:55 -0400559
Jeff Layton3c87ef62015-06-03 16:14:25 -0400560 return rpc_clnt_swap_activate(clnt);
Mel Gormana564b8f2012-07-31 16:45:12 -0700561}
562
563static void nfs_swap_deactivate(struct file *file)
564{
Jeff Laytondad2b012014-09-10 09:03:55 -0400565 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
566
Jeff Layton3c87ef62015-06-03 16:14:25 -0400567 rpc_clnt_swap_deactivate(clnt);
Mel Gormana564b8f2012-07-31 16:45:12 -0700568}
Mel Gormana564b8f2012-07-31 16:45:12 -0700569
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700570const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 .readpage = nfs_readpage,
572 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400573 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 .writepage = nfs_writepage,
575 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700576 .write_begin = nfs_write_begin,
577 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500578 .invalidatepage = nfs_invalidate_page,
579 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 .direct_IO = nfs_direct_IO,
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400581 .migratepage = nfs_migrate_page,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800582 .launder_page = nfs_launder_page,
Mel Gormanf919b192013-07-03 15:02:06 -0700583 .is_dirty_writeback = nfs_check_dirty_writeback,
Andi Kleenf590f332009-09-16 11:50:17 +0200584 .error_remove_page = generic_error_remove_page,
Mel Gormana564b8f2012-07-31 16:45:12 -0700585 .swap_activate = nfs_swap_activate,
586 .swap_deactivate = nfs_swap_deactivate,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587};
588
David Howells6b9b3512009-04-03 16:42:41 +0100589/*
590 * Notification that a PTE pointing to an NFS page is about to be made
591 * writable, implying that someone is about to modify the page through a
592 * shared-writable mapping
593 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700594static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400595{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700596 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400597 struct file *filp = vma->vm_file;
Al Viro6de14722013-09-16 10:53:17 -0400598 struct inode *inode = file_inode(filp);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400599 unsigned pagelen;
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400600 int ret = VM_FAULT_NOPAGE;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700601 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400602
Niels de Vos1e8968c2013-12-17 18:20:16 +0100603 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400604 filp, filp->f_mapping->host->i_ino,
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400605 (long long)page_offset(page));
606
David Howells545db452009-04-03 16:42:44 +0100607 /* make sure the cache has finished storing the page */
Al Viro6de14722013-09-16 10:53:17 -0400608 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
David Howells545db452009-04-03 16:42:44 +0100609
Trond Myklebustef070dc2015-03-03 00:06:35 -0500610 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
611 nfs_wait_bit_killable, TASK_KILLABLE);
612
Trond Myklebust94387fb2007-07-22 17:09:05 -0400613 lock_page(page);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700614 mapping = page_file_mapping(page);
Al Viro6de14722013-09-16 10:53:17 -0400615 if (mapping != inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500616 goto out_unlock;
617
Trond Myklebust2aeb98f2012-01-17 22:04:26 -0500618 wait_on_page_writeback(page);
619
Trond Myklebust94387fb2007-07-22 17:09:05 -0400620 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500621 if (pagelen == 0)
622 goto out_unlock;
623
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400624 ret = VM_FAULT_LOCKED;
625 if (nfs_flush_incompatible(filp, page) == 0 &&
626 nfs_updatepage(filp, page, 0, pagelen) == 0)
627 goto out;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500628
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400629 ret = VM_FAULT_SIGBUS;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500630out_unlock:
Trond Myklebust94387fb2007-07-22 17:09:05 -0400631 unlock_page(page);
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400632out:
633 return ret;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400634}
635
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400636static const struct vm_operations_struct nfs_file_vm_ops = {
Trond Myklebust94387fb2007-07-22 17:09:05 -0400637 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700638 .map_pages = filemap_map_pages,
Trond Myklebust94387fb2007-07-22 17:09:05 -0400639 .page_mkwrite = nfs_vm_page_mkwrite,
640};
641
Trond Myklebust7e94d6c2015-08-17 16:55:18 -0500642static int nfs_need_check_write(struct file *filp, struct inode *inode)
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400643{
644 struct nfs_open_context *ctx;
645
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400646 ctx = nfs_file_open_context(filp);
Andy Adamsondc248262013-08-14 11:59:16 -0400647 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
648 nfs_ctx_key_to_expire(ctx))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400649 return 1;
650 return 0;
651}
652
Al Viroedaf4362014-04-03 14:07:25 -0400653ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
Al Viro6de14722013-09-16 10:53:17 -0400655 struct file *file = iocb->ki_filp;
656 struct inode *inode = file_inode(file);
Chuck Lever7e381172010-02-01 14:17:41 -0500657 unsigned long written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 ssize_t result;
Al Viroedaf4362014-04-03 14:07:25 -0400659 size_t count = iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Al Viro6de14722013-09-16 10:53:17 -0400661 result = nfs_key_timeout_notify(file, inode);
Andy Adamsondc248262013-08-14 11:59:16 -0400662 if (result)
663 return result;
664
Al Viro65a4a1c2015-04-09 14:11:08 -0400665 if (iocb->ki_flags & IOCB_DIRECT) {
666 result = generic_write_checks(iocb, from);
667 if (result <= 0)
668 return result;
669 return nfs_file_direct_write(iocb, from);
670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Al Viro619d30b2014-03-04 21:53:33 -0500672 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
Al Viro65a4a1c2015-04-09 14:11:08 -0400673 file, count, (long long) iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 result = -EBUSY;
676 if (IS_SWAPFILE(inode))
677 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000678 /*
679 * O_APPEND implies that we must revalidate the file length.
680 */
Al Viro2ba48ce2015-04-09 13:52:01 -0400681 if (iocb->ki_flags & IOCB_APPEND) {
Al Viro6de14722013-09-16 10:53:17 -0400682 result = nfs_revalidate_file_size(inode, file);
Trond Myklebust7d52e862005-06-22 17:16:30 +0000683 if (result)
684 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
687 result = count;
688 if (!count)
689 goto out;
690
Al Viroedaf4362014-04-03 14:07:25 -0400691 result = generic_file_write_iter(iocb, from);
Chuck Lever7e381172010-02-01 14:17:41 -0500692 if (result > 0)
693 written = result;
694
Trond Myklebust7e94d6c2015-08-17 16:55:18 -0500695 /* Return error values */
696 if (result >= 0 && nfs_need_check_write(file, inode)) {
Al Viro6de14722013-09-16 10:53:17 -0400697 int err = vfs_fsync(file, 0);
Trond Myklebust200baa22006-12-05 00:35:40 -0500698 if (err < 0)
699 result = err;
700 }
Chuck Lever7e381172010-02-01 14:17:41 -0500701 if (result > 0)
702 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703out:
704 return result;
705
706out_swapfile:
707 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
708 goto out;
709}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400710EXPORT_SYMBOL_GPL(nfs_file_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400712static int
713do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
715 struct inode *inode = filp->f_mapping->host;
716 int status = 0;
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000717 unsigned int saved_type = fl->fl_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Trond Myklebust039c4d72005-10-18 14:20:16 -0700719 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400720 posix_test_lock(filp, fl);
721 if (fl->fl_type != F_UNLCK) {
722 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700723 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 }
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000725 fl->fl_type = saved_type;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700726
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400727 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
Trond Myklebust039c4d72005-10-18 14:20:16 -0700728 goto out_noconflict;
729
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400730 if (is_local)
Trond Myklebust039c4d72005-10-18 14:20:16 -0700731 goto out_noconflict;
732
733 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
734out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700736out_noconflict:
737 fl->fl_type = F_UNLCK;
738 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739}
740
741static int do_vfs_lock(struct file *file, struct file_lock *fl)
742{
Benjamin Coddington4f656362015-10-22 13:38:14 -0400743 return locks_lock_file_wait(file, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400746static int
747do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
749 struct inode *inode = filp->f_mapping->host;
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400750 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 int status;
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 /*
754 * Flush all pending writes before doing anything
755 * with locks..
756 */
Trond Myklebustd9dabc12015-03-26 15:55:49 -0400757 vfs_fsync(filp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400759 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
760 if (!IS_ERR(l_ctx)) {
Benjamin Coddington210c7c12016-01-06 10:40:18 -0500761 status = nfs_iocounter_wait(l_ctx);
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400762 nfs_put_lock_context(l_ctx);
763 if (status < 0)
764 return status;
765 }
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /* NOTE: special case
768 * If we're signalled while cleaning up locks on process exit, we
769 * still need to complete the unlock.
770 */
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400771 /*
772 * Use local locking if mounted with "-onolock" or with appropriate
773 * "-olocal_lock="
774 */
775 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
777 else
778 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return status;
780}
781
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400782static int
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700783is_time_granular(struct timespec *ts) {
784 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
785}
786
787static int
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400788do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
790 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 int status;
792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 /*
794 * Flush all pending writes before doing anything
795 * with locks..
796 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500797 status = nfs_sync_mapping(filp->f_mapping);
798 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 goto out;
800
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400801 /*
802 * Use local locking if mounted with "-onolock" or with appropriate
803 * "-olocal_lock="
804 */
805 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400807 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 if (status < 0)
810 goto out;
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 /*
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700813 * Revalidate the cache if the server has time stamps granular
814 * enough to detect subsecond changes. Otherwise, clear the
815 * cache to prevent missing any changes.
816 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 * This makes locking act as a cache coherency point.
818 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500819 nfs_sync_mapping(filp->f_mapping);
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400820 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700821 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
822 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
823 else
824 nfs_zap_caches(inode);
825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return status;
828}
829
830/*
831 * Lock a (portion of) a file
832 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400833int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400835 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400836 int ret = -ENOLCK;
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400837 int is_local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Al Viro6de14722013-09-16 10:53:17 -0400839 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
840 filp, fl->fl_type, fl->fl_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400842
Chuck Lever91d5b472006-03-20 13:44:14 -0500843 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
845 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700846 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400847 goto out_err;
848
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400849 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
850 is_local = 1;
851
Trond Myklebust21162712008-05-20 19:34:39 -0400852 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
853 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
854 if (ret < 0)
855 goto out_err;
856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 if (IS_GETLK(cmd))
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400859 ret = do_getlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400860 else if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400861 ret = do_unlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400862 else
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400863 ret = do_setlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400864out_err:
865 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400867EXPORT_SYMBOL_GPL(nfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869/*
870 * Lock a (portion of) a file
871 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400872int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873{
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400874 struct inode *inode = filp->f_mapping->host;
875 int is_local = 0;
876
Al Viro6de14722013-09-16 10:53:17 -0400877 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
878 filp, fl->fl_type, fl->fl_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 if (!(fl->fl_flags & FL_FLOCK))
881 return -ENOLCK;
882
Jeff Laytonad0fcd42012-07-23 15:46:23 -0400883 /*
884 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
885 * any standard. In principle we might be able to support LOCK_MAND
886 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
887 * NFS code is not set up for it.
888 */
889 if (fl->fl_type & LOCK_MAND)
890 return -EINVAL;
891
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400892 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
893 is_local = 1;
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 /* We're simulating flock() locks using posix locks on the server */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400897 return do_unlk(filp, cmd, fl, is_local);
898 return do_setlk(filp, cmd, fl, is_local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400900EXPORT_SYMBOL_GPL(nfs_flock);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400901
Jeff Layton04869582011-11-04 13:31:22 -0400902const struct file_operations nfs_file_operations = {
903 .llseek = nfs_file_llseek,
Al Viro3aa2d192014-04-02 20:14:12 -0400904 .read_iter = nfs_file_read,
Al Viroedaf4362014-04-03 14:07:25 -0400905 .write_iter = nfs_file_write,
Jeff Layton04869582011-11-04 13:31:22 -0400906 .mmap = nfs_file_mmap,
907 .open = nfs_file_open,
908 .flush = nfs_file_flush,
909 .release = nfs_file_release,
910 .fsync = nfs_file_fsync,
911 .lock = nfs_lock,
912 .flock = nfs_flock,
913 .splice_read = nfs_file_splice_read,
Al Viro4da54c22014-04-05 04:37:17 -0400914 .splice_write = iter_file_splice_write,
Jeff Layton04869582011-11-04 13:31:22 -0400915 .check_flags = nfs_check_flags,
Jeff Layton1c994a02014-08-27 06:49:41 -0400916 .setlease = simple_nosetlease,
Jeff Layton04869582011-11-04 13:31:22 -0400917};
Bryan Schumakerddda8e02012-07-30 16:05:23 -0400918EXPORT_SYMBOL_GPL(nfs_file_operations);