blob: 8b8d83a526ce2366ae974a87761c9c62c22da7f5 [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);
Trond Myklebust46cb6502008-06-11 16:32:46 -040085 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040087EXPORT_SYMBOL_GPL(nfs_file_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Trond Myklebust980802e2005-06-13 11:14:01 -040089/**
90 * nfs_revalidate_size - Revalidate the file size
91 * @inode - pointer to inode struct
92 * @file - pointer to struct file
93 *
94 * Revalidates the file length. This is basically a wrapper around
95 * nfs_revalidate_inode() that takes into account the fact that we may
96 * have cached writes (in which case we don't care about the server's
97 * idea of what the file length is), or O_DIRECT (in which case we
98 * shouldn't trust the cache).
99 */
100static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101{
102 struct nfs_server *server = NFS_SERVER(inode);
103 struct nfs_inode *nfsi = NFS_I(inode);
104
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400105 if (nfs_have_delegated_attributes(inode))
106 goto out_noreval;
107
Trond Myklebust980802e2005-06-13 11:14:01 -0400108 if (filp->f_flags & O_DIRECT)
109 goto force_reval;
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400110 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
111 goto force_reval;
112 if (nfs_attribute_timeout(inode))
113 goto force_reval;
114out_noreval:
115 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400116force_reval:
117 return __nfs_revalidate_inode(server, inode);
118}
119
Andrew Morton965c8e52012-12-17 15:59:39 -0800120loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
Trond Myklebust980802e2005-06-13 11:14:01 -0400121{
Al Viro6de14722013-09-16 10:53:17 -0400122 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
123 filp, offset, whence);
Chuck Leverb84e06c2008-06-11 17:55:34 -0400124
Josef Bacik06222e42011-07-18 13:21:38 -0400125 /*
Andrew Morton965c8e52012-12-17 15:59:39 -0800126 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
Josef Bacik06222e42011-07-18 13:21:38 -0400127 * the cached file length
128 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800129 if (whence != SEEK_SET && whence != SEEK_CUR) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400130 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400131
Trond Myklebust980802e2005-06-13 11:14:01 -0400132 int retval = nfs_revalidate_file_size(inode, filp);
133 if (retval < 0)
134 return (loff_t)retval;
Andi Kleen79835a712011-09-15 16:06:52 -0700135 }
Trond Myklebustd5e66342008-09-23 17:28:35 -0400136
Andrew Morton965c8e52012-12-17 15:59:39 -0800137 return generic_file_llseek(filp, offset, whence);
Trond Myklebust980802e2005-06-13 11:14:01 -0400138}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400139EXPORT_SYMBOL_GPL(nfs_file_llseek);
Trond Myklebust980802e2005-06-13 11:14:01 -0400140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141/*
142 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400144int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700145nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
Al Viro6de14722013-09-16 10:53:17 -0400147 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Al Viro6de14722013-09-16 10:53:17 -0400149 dprintk("NFS: flush(%pD2)\n", file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Chuck Leverc2459dc2010-02-01 14:17:14 -0500151 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 if ((file->f_mode & FMODE_WRITE) == 0)
153 return 0;
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400154
Trond Myklebust14546c32012-05-06 19:10:59 -0400155 /*
156 * If we're holding a write delegation, then just start the i/o
157 * but don't wait for completion (or send a commit).
158 */
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400159 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebust14546c32012-05-06 19:10:59 -0400160 return filemap_fdatawrite(file->f_mapping);
161
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400162 /* Flush writes to the server and return any errors */
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400163 return vfs_fsync(file, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400165EXPORT_SYMBOL_GPL(nfs_file_flush);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400167ssize_t
Al Viro3aa2d192014-04-02 20:14:12 -0400168nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Al Viro6de14722013-09-16 10:53:17 -0400170 struct inode *inode = file_inode(iocb->ki_filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 ssize_t result;
172
Al Viro2ba48ce2015-04-09 13:52:01 -0400173 if (iocb->ki_flags & IOCB_DIRECT)
Martin K. Petersene19a8a02014-10-14 09:00:44 -0600174 return nfs_file_direct_read(iocb, to, iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Al Viro619d30b2014-03-04 21:53:33 -0500176 dprintk("NFS: read(%pD2, %zu@%lu)\n",
Al Viro6de14722013-09-16 10:53:17 -0400177 iocb->ki_filp,
Al Viro3aa2d192014-04-02 20:14:12 -0400178 iov_iter_count(to), (unsigned long) iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Trond Myklebust874f9462015-03-02 23:32:08 -0500180 result = nfs_revalidate_mapping_protected(inode, iocb->ki_filp->f_mapping);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500181 if (!result) {
Al Viro3aa2d192014-04-02 20:14:12 -0400182 result = generic_file_read_iter(iocb, to);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500183 if (result > 0)
184 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
185 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 return result;
187}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400188EXPORT_SYMBOL_GPL(nfs_file_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400190ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200191nfs_file_splice_read(struct file *filp, loff_t *ppos,
192 struct pipe_inode_info *pipe, size_t count,
193 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
Al Viro6de14722013-09-16 10:53:17 -0400195 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 ssize_t res;
197
Al Viro6de14722013-09-16 10:53:17 -0400198 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
199 filp, (unsigned long) count, (unsigned long long) *ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Trond Myklebust874f9462015-03-02 23:32:08 -0500201 res = nfs_revalidate_mapping_protected(inode, filp->f_mapping);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500202 if (!res) {
Jens Axboef0930ff2007-06-01 11:51:43 +0200203 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500204 if (res > 0)
205 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 return res;
208}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400209EXPORT_SYMBOL_GPL(nfs_file_splice_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400211int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
213{
Al Viro6de14722013-09-16 10:53:17 -0400214 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 int status;
216
Al Viro6de14722013-09-16 10:53:17 -0400217 dprintk("NFS: mmap(%pD2)\n", file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400219 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
220 * so we call that before revalidating the mapping
221 */
222 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400223 if (!status) {
224 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400225 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 return status;
228}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400229EXPORT_SYMBOL_GPL(nfs_file_mmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
231/*
232 * Flush any dirty pages for this process, and check for write errors.
233 * The return status from this call provides a reliable indication of
234 * whether any write errors occurred for this process.
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400235 *
236 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
237 * disk, but it retrieves and clears ctx->error after synching, despite
238 * the two being set at the same time in nfs_context_set_write_error().
239 * This is because the former is used to notify the _next_ call to
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300240 * nfs_file_write() that a write error occurred, and hence cause it to
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400241 * fall back to doing a synchronous write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400243int
Bryan Schumakera5c58892012-06-20 15:53:42 -0400244nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400246 struct nfs_open_context *ctx = nfs_file_open_context(file);
Al Viro6de14722013-09-16 10:53:17 -0400247 struct inode *inode = file_inode(file);
Trond Myklebust05990d12012-09-11 16:01:22 -0400248 int have_error, do_resend, status;
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400249 int ret = 0;
250
Al Viro6de14722013-09-16 10:53:17 -0400251 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Chuck Lever91d5b472006-03-20 13:44:14 -0500253 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebust05990d12012-09-11 16:01:22 -0400254 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400255 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
256 status = nfs_commit_inode(inode, FLUSH_SYNC);
257 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
Trond Myklebust05990d12012-09-11 16:01:22 -0400258 if (have_error) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400259 ret = xchg(&ctx->error, 0);
Trond Myklebust05990d12012-09-11 16:01:22 -0400260 if (ret)
261 goto out;
262 }
263 if (status < 0) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400264 ret = status;
Trond Myklebust05990d12012-09-11 16:01:22 -0400265 goto out;
266 }
267 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
268 if (do_resend)
269 ret = -EAGAIN;
270out:
Bryan Schumakera5c58892012-06-20 15:53:42 -0400271 return ret;
272}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400273EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
Bryan Schumakera5c58892012-06-20 15:53:42 -0400274
275static int
276nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
277{
278 int ret;
Al Viro496ad9a2013-01-23 17:07:38 -0500279 struct inode *inode = file_inode(file);
Bryan Schumakera5c58892012-06-20 15:53:42 -0400280
Trond Myklebustf4ce1292013-08-19 18:59:33 -0400281 trace_nfs_fsync_enter(inode);
282
Trond Myklebusta0815d52015-03-25 18:58:53 -0400283 nfs_inode_dio_wait(inode);
Trond Myklebust05990d12012-09-11 16:01:22 -0400284 do {
285 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
286 if (ret != 0)
287 break;
288 mutex_lock(&inode->i_mutex);
289 ret = nfs_file_fsync_commit(file, start, end, datasync);
290 mutex_unlock(&inode->i_mutex);
Trond Myklebustdcfc4f22012-09-11 16:19:38 -0400291 /*
292 * If nfs_file_fsync_commit detected a server reboot, then
293 * resend all dirty pages that might have been covered by
294 * the NFS_CONTEXT_RESEND_WRITES flag
295 */
296 start = 0;
297 end = LLONG_MAX;
Trond Myklebust05990d12012-09-11 16:01:22 -0400298 } while (ret == -EAGAIN);
299
Trond Myklebustf4ce1292013-08-19 18:59:33 -0400300 trace_nfs_fsync_exit(inode, ret);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400301 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303
304/*
Peter Staubach38c73042009-08-10 08:54:16 -0400305 * Decide whether a read/modify/write cycle may be more efficient
306 * then a modify/write/read cycle when writing to a page in the
307 * page cache.
308 *
309 * The modify/write/read cycle may occur if a page is read before
310 * being completely filled by the writer. In this situation, the
311 * page must be completely written to stable storage on the server
312 * before it can be refilled by reading in the page from the server.
313 * This can lead to expensive, small, FILE_SYNC mode writes being
314 * done.
315 *
316 * It may be more efficient to read the page first if the file is
317 * open for reading in addition to writing, the page is not marked
318 * as Uptodate, it is not dirty or waiting to be committed,
319 * indicating that it was previously allocated and then modified,
320 * that there were valid bytes of data in that range of the file,
321 * and that the new data won't completely replace the old data in
322 * that range of the file.
323 */
324static int nfs_want_read_modify_write(struct file *file, struct page *page,
325 loff_t pos, unsigned len)
326{
327 unsigned int pglen = nfs_page_length(page);
328 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
329 unsigned int end = offset + len;
330
Christoph Hellwig612aa982014-09-10 08:23:30 -0700331 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
332 if (!PageUptodate(page))
333 return 1;
334 return 0;
335 }
336
Peter Staubach38c73042009-08-10 08:54:16 -0400337 if ((file->f_mode & FMODE_READ) && /* open for read? */
338 !PageUptodate(page) && /* Uptodate? */
339 !PagePrivate(page) && /* i/o request already? */
340 pglen && /* valid bytes of file? */
341 (end < pglen || offset)) /* replace all valid bytes? */
342 return 1;
343 return 0;
344}
345
346/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700347 * This does the "real" work of the write. We must allocate and lock the
348 * page to be sent back to the generic routine, which then copies the
349 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 *
351 * If the writer ends up delaying the write, the writer needs to
352 * increment the page use counts until he is done with the page.
353 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700354static int nfs_write_begin(struct file *file, struct address_space *mapping,
355 loff_t pos, unsigned len, unsigned flags,
356 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700358 int ret;
Peter Staubach38c73042009-08-10 08:54:16 -0400359 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700360 struct page *page;
Peter Staubach38c73042009-08-10 08:54:16 -0400361 int once_thru = 0;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700362
Niels de Vos1e8968c2013-12-17 18:20:16 +0100363 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400364 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400365
Peter Staubach38c73042009-08-10 08:54:16 -0400366start:
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400367 /*
368 * Prevent starvation issues if someone is doing a consistency
369 * sync-to-disk
370 */
NeilBrown74316202014-07-07 15:16:04 +1000371 ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
372 nfs_wait_bit_killable, TASK_KILLABLE);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400373 if (ret)
374 return ret;
Trond Myklebustaa5accea2015-02-27 22:54:19 -0500375 /*
376 * Wait for O_DIRECT to complete
377 */
378 nfs_inode_dio_wait(mapping->host);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400379
Nick Piggin54566b22009-01-04 12:00:53 -0800380 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700381 if (!page)
382 return -ENOMEM;
383 *pagep = page;
384
385 ret = nfs_flush_incompatible(file, page);
386 if (ret) {
387 unlock_page(page);
388 page_cache_release(page);
Peter Staubach38c73042009-08-10 08:54:16 -0400389 } else if (!once_thru &&
390 nfs_want_read_modify_write(file, page, pos, len)) {
391 once_thru = 1;
392 ret = nfs_readpage(file, page);
393 page_cache_release(page);
394 if (!ret)
395 goto start;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700396 }
397 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399
Nick Piggin4899f9c2007-10-16 01:25:16 -0700400static int nfs_write_end(struct file *file, struct address_space *mapping,
401 loff_t pos, unsigned len, unsigned copied,
402 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700404 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
Andy Adamsondc248262013-08-14 11:59:16 -0400405 struct nfs_open_context *ctx = nfs_file_open_context(file);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700406 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Niels de Vos1e8968c2013-12-17 18:20:16 +0100408 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400409 file, mapping->host->i_ino, len, (long long) pos);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400410
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400411 /*
412 * Zero any uninitialised parts of the page, and then mark the page
413 * as up to date if it turns out that we're extending the file.
414 */
415 if (!PageUptodate(page)) {
416 unsigned pglen = nfs_page_length(page);
417 unsigned end = offset + len;
418
419 if (pglen == 0) {
420 zero_user_segments(page, 0, offset,
421 end, PAGE_CACHE_SIZE);
422 SetPageUptodate(page);
423 } else if (end >= pglen) {
424 zero_user_segment(page, end, PAGE_CACHE_SIZE);
425 if (offset == 0)
426 SetPageUptodate(page);
427 } else
428 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
429 }
430
Nick Piggin4899f9c2007-10-16 01:25:16 -0700431 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700432
433 unlock_page(page);
434 page_cache_release(page);
435
Chuck Lever3d509e52007-12-20 14:55:04 -0500436 if (status < 0)
437 return status;
Andy Adamson2701d082012-05-24 13:13:24 -0400438 NFS_I(mapping->host)->write_io += copied;
Andy Adamsondc248262013-08-14 11:59:16 -0400439
440 if (nfs_ctx_key_to_expire(ctx)) {
441 status = nfs_wb_all(mapping->host);
442 if (status < 0)
443 return status;
444 }
445
Chuck Lever3d509e52007-12-20 14:55:04 -0500446 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447}
448
David Howells6b9b3512009-04-03 16:42:41 +0100449/*
450 * Partially or wholly invalidate a page
451 * - Release the private state associated with a page if undergoing complete
452 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100453 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100454 * - Caller holds page lock
455 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400456static void nfs_invalidate_page(struct page *page, unsigned int offset,
457 unsigned int length)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500458{
Lukas Czernerd47992f2013-05-21 23:17:23 -0400459 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
460 page, offset, length);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400461
Lukas Czernerd47992f2013-05-21 23:17:23 -0400462 if (offset != 0 || length < PAGE_CACHE_SIZE)
Trond Myklebust1c759502006-10-09 16:18:38 -0400463 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400464 /* Cancel any unstarted writes on this page */
Mel Gormand56b4dd2012-07-31 16:45:06 -0700465 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
David Howells545db452009-04-03 16:42:44 +0100466
467 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500468}
469
David Howells6b9b3512009-04-03 16:42:41 +0100470/*
471 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100472 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100473 * - Caller holds page lock
474 * - Return true (may release page) or false (may not)
475 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500476static int nfs_release_page(struct page *page, gfp_t gfp)
477{
Trond Myklebustb608b282010-07-30 15:31:54 -0400478 struct address_space *mapping = page->mapping;
479
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400480 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
481
NeilBrown95905442014-09-24 11:28:32 +1000482 /* Always try to initiate a 'commit' if relevant, but only
NeilBrown1aff5252014-09-24 11:28:32 +1000483 * wait for it if __GFP_WAIT is set. Even then, only wait 1
484 * second and only if the 'bdi' is not congested.
NeilBrown95905442014-09-24 11:28:32 +1000485 * Waiting indefinitely can cause deadlocks when the NFS
NeilBrown1aff5252014-09-24 11:28:32 +1000486 * server is on this machine, when a new TCP connection is
487 * needed and in other rare cases. There is no particular
488 * need to wait extensively here. A short wait has the
489 * benefit that someone else can worry about the freezer.
Jeff Layton5cf02d02012-07-23 13:58:51 -0400490 */
NeilBrown95905442014-09-24 11:28:32 +1000491 if (mapping) {
492 struct nfs_server *nfss = NFS_SERVER(mapping->host);
493 nfs_commit_inode(mapping->host, 0);
494 if ((gfp & __GFP_WAIT) &&
NeilBrown353db792014-09-24 11:28:32 +1000495 !bdi_write_congested(&nfss->backing_dev_info)) {
NeilBrown95905442014-09-24 11:28:32 +1000496 wait_on_page_bit_killable_timeout(page, PG_private,
497 HZ);
NeilBrown353db792014-09-24 11:28:32 +1000498 if (PagePrivate(page))
499 set_bdi_congested(&nfss->backing_dev_info,
500 BLK_RW_ASYNC);
NeilBrown95905442014-09-24 11:28:32 +1000501 }
Trond Myklebustb608b282010-07-30 15:31:54 -0400502 }
Trond Myklebuste3db7692007-01-10 23:15:39 -0800503 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100504 if (PagePrivate(page))
505 return 0;
506 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800507}
508
Mel Gormanf919b192013-07-03 15:02:06 -0700509static void nfs_check_dirty_writeback(struct page *page,
510 bool *dirty, bool *writeback)
511{
512 struct nfs_inode *nfsi;
513 struct address_space *mapping = page_file_mapping(page);
514
515 if (!mapping || PageSwapCache(page))
516 return;
517
518 /*
519 * Check if an unstable page is currently being committed and
520 * if so, have the VM treat it as if the page is under writeback
521 * so it will not block due to pages that will shortly be freeable.
522 */
523 nfsi = NFS_I(mapping->host);
524 if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
525 *writeback = true;
526 return;
527 }
528
529 /*
530 * If PagePrivate() is set, then the page is not freeable and as the
531 * inode is not being committed, it's not going to be cleaned in the
532 * near future so treat it as dirty
533 */
534 if (PagePrivate(page))
535 *dirty = true;
536}
537
David Howells6b9b3512009-04-03 16:42:41 +0100538/*
539 * Attempt to clear the private state associated with a page when an error
540 * occurs that requires the cached contents of an inode to be written back or
541 * destroyed
David Howells545db452009-04-03 16:42:44 +0100542 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100543 * - Caller holds page lock
544 * - Return 0 if successful, -error otherwise
545 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800546static int nfs_launder_page(struct page *page)
547{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700548 struct inode *inode = page_file_mapping(page)->host;
David Howells545db452009-04-03 16:42:44 +0100549 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400550
551 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
552 inode->i_ino, (long long)page_offset(page));
553
David Howells545db452009-04-03 16:42:44 +0100554 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400555 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500556}
557
Mel Gormana564b8f2012-07-31 16:45:12 -0700558#ifdef CONFIG_NFS_SWAP
559static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
560 sector_t *span)
561{
Jeff Laytondad2b012014-09-10 09:03:55 -0400562 int ret;
563 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
564
Mel Gormana564b8f2012-07-31 16:45:12 -0700565 *span = sis->pages;
Jeff Laytondad2b012014-09-10 09:03:55 -0400566
567 rcu_read_lock();
568 ret = xs_swapper(rcu_dereference(clnt->cl_xprt), 1);
569 rcu_read_unlock();
570
571 return ret;
Mel Gormana564b8f2012-07-31 16:45:12 -0700572}
573
574static void nfs_swap_deactivate(struct file *file)
575{
Jeff Laytondad2b012014-09-10 09:03:55 -0400576 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
577
578 rcu_read_lock();
579 xs_swapper(rcu_dereference(clnt->cl_xprt), 0);
580 rcu_read_unlock();
Mel Gormana564b8f2012-07-31 16:45:12 -0700581}
582#endif
583
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700584const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 .readpage = nfs_readpage,
586 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400587 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 .writepage = nfs_writepage,
589 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700590 .write_begin = nfs_write_begin,
591 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500592 .invalidatepage = nfs_invalidate_page,
593 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 .direct_IO = nfs_direct_IO,
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400595 .migratepage = nfs_migrate_page,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800596 .launder_page = nfs_launder_page,
Mel Gormanf919b192013-07-03 15:02:06 -0700597 .is_dirty_writeback = nfs_check_dirty_writeback,
Andi Kleenf590f332009-09-16 11:50:17 +0200598 .error_remove_page = generic_error_remove_page,
Mel Gormana564b8f2012-07-31 16:45:12 -0700599#ifdef CONFIG_NFS_SWAP
600 .swap_activate = nfs_swap_activate,
601 .swap_deactivate = nfs_swap_deactivate,
602#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603};
604
David Howells6b9b3512009-04-03 16:42:41 +0100605/*
606 * Notification that a PTE pointing to an NFS page is about to be made
607 * writable, implying that someone is about to modify the page through a
608 * shared-writable mapping
609 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700610static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400611{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700612 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400613 struct file *filp = vma->vm_file;
Al Viro6de14722013-09-16 10:53:17 -0400614 struct inode *inode = file_inode(filp);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400615 unsigned pagelen;
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400616 int ret = VM_FAULT_NOPAGE;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700617 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400618
Niels de Vos1e8968c2013-12-17 18:20:16 +0100619 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
Al Viro6de14722013-09-16 10:53:17 -0400620 filp, filp->f_mapping->host->i_ino,
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400621 (long long)page_offset(page));
622
David Howells545db452009-04-03 16:42:44 +0100623 /* make sure the cache has finished storing the page */
Al Viro6de14722013-09-16 10:53:17 -0400624 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
David Howells545db452009-04-03 16:42:44 +0100625
Trond Myklebustef070dc2015-03-03 00:06:35 -0500626 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
627 nfs_wait_bit_killable, TASK_KILLABLE);
628
Trond Myklebust94387fb2007-07-22 17:09:05 -0400629 lock_page(page);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700630 mapping = page_file_mapping(page);
Al Viro6de14722013-09-16 10:53:17 -0400631 if (mapping != inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500632 goto out_unlock;
633
Trond Myklebust2aeb98f2012-01-17 22:04:26 -0500634 wait_on_page_writeback(page);
635
Trond Myklebust94387fb2007-07-22 17:09:05 -0400636 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500637 if (pagelen == 0)
638 goto out_unlock;
639
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400640 ret = VM_FAULT_LOCKED;
641 if (nfs_flush_incompatible(filp, page) == 0 &&
642 nfs_updatepage(filp, page, 0, pagelen) == 0)
643 goto out;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500644
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400645 ret = VM_FAULT_SIGBUS;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500646out_unlock:
Trond Myklebust94387fb2007-07-22 17:09:05 -0400647 unlock_page(page);
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400648out:
649 return ret;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400650}
651
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400652static const struct vm_operations_struct nfs_file_vm_ops = {
Trond Myklebust94387fb2007-07-22 17:09:05 -0400653 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -0700654 .map_pages = filemap_map_pages,
Trond Myklebust94387fb2007-07-22 17:09:05 -0400655 .page_mkwrite = nfs_vm_page_mkwrite,
656};
657
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400658static int nfs_need_sync_write(struct file *filp, struct inode *inode)
659{
660 struct nfs_open_context *ctx;
661
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100662 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400663 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400664 ctx = nfs_file_open_context(filp);
Andy Adamsondc248262013-08-14 11:59:16 -0400665 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
666 nfs_ctx_key_to_expire(ctx))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400667 return 1;
668 return 0;
669}
670
Al Viroedaf4362014-04-03 14:07:25 -0400671ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
Al Viro6de14722013-09-16 10:53:17 -0400673 struct file *file = iocb->ki_filp;
674 struct inode *inode = file_inode(file);
Chuck Lever7e381172010-02-01 14:17:41 -0500675 unsigned long written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 ssize_t result;
Al Viroedaf4362014-04-03 14:07:25 -0400677 size_t count = iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Al Viro6de14722013-09-16 10:53:17 -0400679 result = nfs_key_timeout_notify(file, inode);
Andy Adamsondc248262013-08-14 11:59:16 -0400680 if (result)
681 return result;
682
Al Viro65a4a1c2015-04-09 14:11:08 -0400683 if (iocb->ki_flags & IOCB_DIRECT) {
684 result = generic_write_checks(iocb, from);
685 if (result <= 0)
686 return result;
687 return nfs_file_direct_write(iocb, from);
688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Al Viro619d30b2014-03-04 21:53:33 -0500690 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
Al Viro65a4a1c2015-04-09 14:11:08 -0400691 file, count, (long long) iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 result = -EBUSY;
694 if (IS_SWAPFILE(inode))
695 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000696 /*
697 * O_APPEND implies that we must revalidate the file length.
698 */
Al Viro2ba48ce2015-04-09 13:52:01 -0400699 if (iocb->ki_flags & IOCB_APPEND) {
Al Viro6de14722013-09-16 10:53:17 -0400700 result = nfs_revalidate_file_size(inode, file);
Trond Myklebust7d52e862005-06-22 17:16:30 +0000701 if (result)
702 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
705 result = count;
706 if (!count)
707 goto out;
708
Al Viroedaf4362014-04-03 14:07:25 -0400709 result = generic_file_write_iter(iocb, from);
Chuck Lever7e381172010-02-01 14:17:41 -0500710 if (result > 0)
711 written = result;
712
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100713 /* Return error values for O_DSYNC and IS_SYNC() */
Al Viro6de14722013-09-16 10:53:17 -0400714 if (result >= 0 && nfs_need_sync_write(file, inode)) {
715 int err = vfs_fsync(file, 0);
Trond Myklebust200baa22006-12-05 00:35:40 -0500716 if (err < 0)
717 result = err;
718 }
Chuck Lever7e381172010-02-01 14:17:41 -0500719 if (result > 0)
720 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721out:
722 return result;
723
724out_swapfile:
725 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
726 goto out;
727}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400728EXPORT_SYMBOL_GPL(nfs_file_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400730static int
731do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 struct inode *inode = filp->f_mapping->host;
734 int status = 0;
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000735 unsigned int saved_type = fl->fl_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Trond Myklebust039c4d72005-10-18 14:20:16 -0700737 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400738 posix_test_lock(filp, fl);
739 if (fl->fl_type != F_UNLCK) {
740 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700741 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000743 fl->fl_type = saved_type;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700744
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400745 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
Trond Myklebust039c4d72005-10-18 14:20:16 -0700746 goto out_noconflict;
747
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400748 if (is_local)
Trond Myklebust039c4d72005-10-18 14:20:16 -0700749 goto out_noconflict;
750
751 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
752out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700754out_noconflict:
755 fl->fl_type = F_UNLCK;
756 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
758
759static int do_vfs_lock(struct file *file, struct file_lock *fl)
760{
761 int res = 0;
762 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
763 case FL_POSIX:
764 res = posix_lock_file_wait(file, fl);
765 break;
766 case FL_FLOCK:
767 res = flock_lock_file_wait(file, fl);
768 break;
769 default:
770 BUG();
771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 return res;
773}
774
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400775static int
776do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
778 struct inode *inode = filp->f_mapping->host;
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400779 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 int status;
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 /*
783 * Flush all pending writes before doing anything
784 * with locks..
785 */
Trond Myklebustd9dabc12015-03-26 15:55:49 -0400786 vfs_fsync(filp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
Trond Myklebust7a8203d2013-04-08 21:49:53 -0400788 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
789 if (!IS_ERR(l_ctx)) {
790 status = nfs_iocounter_wait(&l_ctx->io_count);
791 nfs_put_lock_context(l_ctx);
792 if (status < 0)
793 return status;
794 }
795
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 /* NOTE: special case
797 * If we're signalled while cleaning up locks on process exit, we
798 * still need to complete the unlock.
799 */
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400800 /*
801 * Use local locking if mounted with "-onolock" or with appropriate
802 * "-olocal_lock="
803 */
804 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
806 else
807 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 return status;
809}
810
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400811static int
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700812is_time_granular(struct timespec *ts) {
813 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
814}
815
816static int
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400817do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
819 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 int status;
821
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 /*
823 * Flush all pending writes before doing anything
824 * with locks..
825 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500826 status = nfs_sync_mapping(filp->f_mapping);
827 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 goto out;
829
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400830 /*
831 * Use local locking if mounted with "-onolock" or with appropriate
832 * "-olocal_lock="
833 */
834 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400836 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 if (status < 0)
839 goto out;
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 /*
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700842 * Revalidate the cache if the server has time stamps granular
843 * enough to detect subsecond changes. Otherwise, clear the
844 * cache to prevent missing any changes.
845 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 * This makes locking act as a cache coherency point.
847 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500848 nfs_sync_mapping(filp->f_mapping);
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400849 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700850 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
851 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
852 else
853 nfs_zap_caches(inode);
854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 return status;
857}
858
859/*
860 * Lock a (portion of) a file
861 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400862int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400864 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400865 int ret = -ENOLCK;
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400866 int is_local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Al Viro6de14722013-09-16 10:53:17 -0400868 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
869 filp, fl->fl_type, fl->fl_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400871
Chuck Lever91d5b472006-03-20 13:44:14 -0500872 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
874 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700875 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400876 goto out_err;
877
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400878 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
879 is_local = 1;
880
Trond Myklebust21162712008-05-20 19:34:39 -0400881 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
882 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
883 if (ret < 0)
884 goto out_err;
885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 if (IS_GETLK(cmd))
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400888 ret = do_getlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400889 else if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400890 ret = do_unlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400891 else
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400892 ret = do_setlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400893out_err:
894 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400896EXPORT_SYMBOL_GPL(nfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898/*
899 * Lock a (portion of) a file
900 */
Bryan Schumakerce4ef7c2012-07-16 16:39:15 -0400901int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400903 struct inode *inode = filp->f_mapping->host;
904 int is_local = 0;
905
Al Viro6de14722013-09-16 10:53:17 -0400906 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
907 filp, fl->fl_type, fl->fl_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (!(fl->fl_flags & FL_FLOCK))
910 return -ENOLCK;
911
Jeff Laytonad0fcd42012-07-23 15:46:23 -0400912 /*
913 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
914 * any standard. In principle we might be able to support LOCK_MAND
915 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
916 * NFS code is not set up for it.
917 */
918 if (fl->fl_type & LOCK_MAND)
919 return -EINVAL;
920
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400921 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
922 is_local = 1;
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 /* We're simulating flock() locks using posix locks on the server */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400926 return do_unlk(filp, cmd, fl, is_local);
927 return do_setlk(filp, cmd, fl, is_local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400929EXPORT_SYMBOL_GPL(nfs_flock);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400930
Jeff Layton04869582011-11-04 13:31:22 -0400931const struct file_operations nfs_file_operations = {
932 .llseek = nfs_file_llseek,
Al Viro3aa2d192014-04-02 20:14:12 -0400933 .read_iter = nfs_file_read,
Al Viroedaf4362014-04-03 14:07:25 -0400934 .write_iter = nfs_file_write,
Jeff Layton04869582011-11-04 13:31:22 -0400935 .mmap = nfs_file_mmap,
936 .open = nfs_file_open,
937 .flush = nfs_file_flush,
938 .release = nfs_file_release,
939 .fsync = nfs_file_fsync,
940 .lock = nfs_lock,
941 .flock = nfs_flock,
942 .splice_read = nfs_file_splice_read,
Al Viro4da54c22014-04-05 04:37:17 -0400943 .splice_write = iter_file_splice_write,
Jeff Layton04869582011-11-04 13:31:22 -0400944 .check_flags = nfs_check_flags,
Jeff Layton1c994a02014-08-27 06:49:41 -0400945 .setlease = simple_nosetlease,
Jeff Layton04869582011-11-04 13:31:22 -0400946};
Bryan Schumakerddda8e02012-07-30 16:05:23 -0400947EXPORT_SYMBOL_GPL(nfs_file_operations);