NFS: inter ssc open
[linux-block.git] / fs / nfs / nfs4file.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
ce4ef7c0
BS
2/*
3 * linux/fs/nfs/file.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 */
f4ac1674 7#include <linux/fs.h>
bea51b30 8#include <linux/file.h>
f4ac1674 9#include <linux/falloc.h>
ce4ef7c0 10#include <linux/nfs_fs.h>
ec4b0925 11#include <linux/file.h>
5445b1fb 12#include "delegation.h"
ce4ef7c0 13#include "internal.h"
5445b1fb 14#include "iostat.h"
a4ff1468 15#include "fscache.h"
ce4ef7c0
BS
16#include "pnfs.h"
17
81b79afb
TM
18#include "nfstrace.h"
19
1c6dcbe5
AS
20#ifdef CONFIG_NFS_V4_2
21#include "nfs42.h"
22#endif
23
ce4ef7c0
BS
24#define NFSDBG_FACILITY NFSDBG_FILE
25
26static int
27nfs4_file_open(struct inode *inode, struct file *filp)
28{
29 struct nfs_open_context *ctx;
be62a1a8 30 struct dentry *dentry = file_dentry(filp);
ce4ef7c0
BS
31 struct dentry *parent = NULL;
32 struct inode *dir;
33 unsigned openflags = filp->f_flags;
34 struct iattr attr;
35 int err;
36
ce4ef7c0
BS
37 /*
38 * If no cached dentry exists or if it's negative, NFSv4 handled the
39 * opens in ->lookup() or ->create().
40 *
41 * We only get this far for a cached positive dentry. We skipped
42 * revalidation, so handle it here by dropping the dentry and returning
43 * -EOPENSTALE. The VFS will retry the lookup/create/open.
44 */
45
6de1472f 46 dprintk("NFS: open file(%pd2)\n", dentry);
ce4ef7c0 47
18a60089
BC
48 err = nfs_check_flags(openflags);
49 if (err)
50 return err;
51
ce4ef7c0 52 if ((openflags & O_ACCMODE) == 3)
44942b4e 53 return nfs_open(inode, filp);
ce4ef7c0
BS
54
55 /* We can't create new files here */
56 openflags &= ~(O_CREAT|O_EXCL);
57
58 parent = dget_parent(dentry);
2b0143b5 59 dir = d_inode(parent);
ce4ef7c0 60
532d4def 61 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
ce4ef7c0
BS
62 err = PTR_ERR(ctx);
63 if (IS_ERR(ctx))
64 goto out;
65
66 attr.ia_valid = ATTR_OPEN;
67 if (openflags & O_TRUNC) {
68 attr.ia_valid |= ATTR_SIZE;
69 attr.ia_size = 0;
8b7d9d09 70 filemap_write_and_wait(inode->i_mapping);
ce4ef7c0
BS
71 }
72
c5c3fb5f 73 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
ce4ef7c0
BS
74 if (IS_ERR(inode)) {
75 err = PTR_ERR(inode);
76 switch (err) {
ce4ef7c0 77 default:
90cf500e
TM
78 goto out_put_ctx;
79 case -ENOENT:
80 case -ESTALE:
81 case -EISDIR:
82 case -ENOTDIR:
83 case -ELOOP:
ce4ef7c0
BS
84 goto out_drop;
85 }
86 }
2b0143b5 87 if (inode != d_inode(dentry))
ce4ef7c0
BS
88 goto out_drop;
89
90 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
91 nfs_file_set_open_context(filp, ctx);
f1fe29b4 92 nfs_fscache_open_file(inode, filp);
ce4ef7c0
BS
93 err = 0;
94
95out_put_ctx:
96 put_nfs_open_context(ctx);
97out:
98 dput(parent);
99 return err;
100
101out_drop:
102 d_drop(dentry);
103 err = -EOPENSTALE;
104 goto out_put_ctx;
105}
106
5445b1fb
TM
107/*
108 * Flush all dirty pages, and check for write errors.
109 */
110static int
111nfs4_file_flush(struct file *file, fl_owner_t id)
112{
113 struct inode *inode = file_inode(file);
114
115 dprintk("NFS: flush(%pD2)\n", file);
116
117 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
118 if ((file->f_mode & FMODE_WRITE) == 0)
119 return 0;
120
121 /*
122 * If we're holding a write delegation, then check if we're required
123 * to flush the i/o on close. If not, then just start the i/o now.
124 */
125 if (!nfs4_delegation_flush_on_close(inode))
126 return filemap_fdatawrite(file->f_mapping);
127
128 /* Flush writes to the server and return any errors */
aded8d7b 129 return nfs_wb_all(inode);
5445b1fb
TM
130}
131
1c6dcbe5 132#ifdef CONFIG_NFS_V4_2
64bf5ff5
DC
133static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
134 struct file *file_out, loff_t pos_out,
135 size_t count, unsigned int flags)
2e72448b 136{
0491567b 137 struct nfs42_copy_notify_res *cn_resp = NULL;
1d38f3f0
OK
138 struct nl4_server *nss = NULL;
139 nfs4_stateid *cnrs = NULL;
0491567b
OK
140 ssize_t ret;
141
5dae222a
AG
142 /* Only offload copy if superblock is the same */
143 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
144 return -EXDEV;
0769663b
OK
145 if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY))
146 return -EOPNOTSUPP;
837bb1d7 147 if (file_inode(file_in) == file_inode(file_out))
0769663b 148 return -EOPNOTSUPP;
0491567b
OK
149 if (!nfs42_files_from_same_server(file_in, file_out)) {
150 cn_resp = kzalloc(sizeof(struct nfs42_copy_notify_res),
151 GFP_NOFS);
152 if (unlikely(cn_resp == NULL))
153 return -ENOMEM;
154
155 ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp);
156 if (ret) {
157 ret = -EOPNOTSUPP;
158 goto out;
159 }
1d38f3f0
OK
160 nss = &cn_resp->cnr_src;
161 cnrs = &cn_resp->cnr_stateid;
0491567b 162 }
1d38f3f0
OK
163 ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count,
164 nss, cnrs);
0491567b
OK
165out:
166 kfree(cn_resp);
167 return ret;
2e72448b
AS
168}
169
64bf5ff5
DC
170static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
171 struct file *file_out, loff_t pos_out,
172 size_t count, unsigned int flags)
173{
174 ssize_t ret;
175
176 ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
177 flags);
5dae222a 178 if (ret == -EOPNOTSUPP || ret == -EXDEV)
64bf5ff5
DC
179 ret = generic_copy_file_range(file_in, pos_in, file_out,
180 pos_out, count, flags);
181 return ret;
182}
183
1c6dcbe5
AS
184static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
185{
186 loff_t ret;
187
188 switch (whence) {
189 case SEEK_HOLE:
190 case SEEK_DATA:
191 ret = nfs42_proc_llseek(filep, offset, whence);
192 if (ret != -ENOTSUPP)
193 return ret;
01e03bdc 194 /* Fall through */
1c6dcbe5
AS
195 default:
196 return nfs_file_llseek(filep, offset, whence);
197 }
198}
f4ac1674
AS
199
200static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
201{
202 struct inode *inode = file_inode(filep);
203 long ret;
204
205 if (!S_ISREG(inode->i_mode))
206 return -EOPNOTSUPP;
207
624bd5b7 208 if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
f4ac1674
AS
209 return -EOPNOTSUPP;
210
211 ret = inode_newsize_ok(inode, offset + len);
212 if (ret < 0)
213 return ret;
214
624bd5b7 215 if (mode & FALLOC_FL_PUNCH_HOLE)
f830f7dd
AS
216 return nfs42_proc_deallocate(filep, offset, len);
217 return nfs42_proc_allocate(filep, offset, len);
f4ac1674 218}
bea51b30 219
42ec3d4c
DW
220static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
221 struct file *dst_file, loff_t dst_off, loff_t count,
2e5dfc99 222 unsigned int remap_flags)
bea51b30
PT
223{
224 struct inode *dst_inode = file_inode(dst_file);
811b7b85 225 struct nfs_server *server = NFS_SERVER(dst_inode);
04b38d60 226 struct inode *src_inode = file_inode(src_file);
811b7b85 227 unsigned int bs = server->clone_blksize;
21fad313 228 bool same_inode = false;
bea51b30
PT
229 int ret;
230
9026b3a9
DW
231 /* NFS does not support deduplication. */
232 if (remap_flags & REMAP_FILE_DEDUP)
233 return -EOPNOTSUPP;
234
235 if (remap_flags & ~REMAP_FILE_ADVISORY)
2e5dfc99
DW
236 return -EINVAL;
237
811b7b85
PT
238 /* check alignment w.r.t. clone_blksize */
239 ret = -EINVAL;
240 if (bs) {
241 if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
04b38d60 242 goto out;
811b7b85 243 if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
04b38d60 244 goto out;
811b7b85
PT
245 }
246
04b38d60
CH
247 if (src_inode == dst_inode)
248 same_inode = true;
21fad313 249
bea51b30 250 /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
21fad313 251 if (same_inode) {
5955102c 252 inode_lock(src_inode);
21fad313 253 } else if (dst_inode < src_inode) {
5955102c
AV
254 inode_lock_nested(dst_inode, I_MUTEX_PARENT);
255 inode_lock_nested(src_inode, I_MUTEX_CHILD);
bea51b30 256 } else {
5955102c
AV
257 inode_lock_nested(src_inode, I_MUTEX_PARENT);
258 inode_lock_nested(dst_inode, I_MUTEX_CHILD);
bea51b30
PT
259 }
260
261 /* flush all pending writes on both src and dst so that server
262 * has the latest data */
263 ret = nfs_sync_inode(src_inode);
264 if (ret)
265 goto out_unlock;
266 ret = nfs_sync_inode(dst_inode);
267 if (ret)
268 goto out_unlock;
269
04b38d60 270 ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
bea51b30
PT
271
272 /* truncate inode page cache of the dst range so that future reads can fetch
273 * new data from server */
274 if (!ret)
275 truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
276
277out_unlock:
21fad313 278 if (same_inode) {
5955102c 279 inode_unlock(src_inode);
21fad313 280 } else if (dst_inode < src_inode) {
5955102c
AV
281 inode_unlock(src_inode);
282 inode_unlock(dst_inode);
bea51b30 283 } else {
5955102c
AV
284 inode_unlock(dst_inode);
285 inode_unlock(src_inode);
bea51b30 286 }
04b38d60 287out:
42ec3d4c 288 return ret < 0 ? ret : count;
bea51b30 289}
ec4b0925
OK
290
291static int read_name_gen = 1;
292#define SSC_READ_NAME_BODY "ssc_read_%d"
293
294struct file *
295nfs42_ssc_open(struct vfsmount *ss_mnt, struct nfs_fh *src_fh,
296 nfs4_stateid *stateid)
297{
298 struct nfs_fattr fattr;
299 struct file *filep, *res;
300 struct nfs_server *server;
301 struct inode *r_ino = NULL;
302 struct nfs_open_context *ctx;
303 struct nfs4_state_owner *sp;
304 char *read_name;
305 int len, status = 0;
306
307 server = NFS_SERVER(ss_mnt->mnt_root->d_inode);
308
309 nfs_fattr_init(&fattr);
310
311 status = nfs4_proc_getattr(server, src_fh, &fattr, NULL, NULL);
312 if (status < 0) {
313 res = ERR_PTR(status);
314 goto out;
315 }
316
317 res = ERR_PTR(-ENOMEM);
318 len = strlen(SSC_READ_NAME_BODY) + 16;
319 read_name = kzalloc(len, GFP_NOFS);
320 if (read_name == NULL)
321 goto out;
322 snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
323
324 r_ino = nfs_fhget(ss_mnt->mnt_root->d_inode->i_sb, src_fh, &fattr,
325 NULL);
326 if (IS_ERR(r_ino)) {
327 res = ERR_CAST(r_ino);
328 goto out;
329 }
330
331 filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, FMODE_READ,
332 r_ino->i_fop);
333 if (IS_ERR(filep)) {
334 res = ERR_CAST(filep);
335 goto out;
336 }
337 filep->f_mode |= FMODE_READ;
338
339 ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode,
340 filep);
341 if (IS_ERR(ctx)) {
342 res = ERR_CAST(ctx);
343 goto out_filep;
344 }
345
346 res = ERR_PTR(-EINVAL);
347 sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
348 if (sp == NULL)
349 goto out_ctx;
350
351 ctx->state = nfs4_get_open_state(r_ino, sp);
352 if (ctx->state == NULL)
353 goto out_stateowner;
354
355 set_bit(NFS_OPEN_STATE, &ctx->state->flags);
356 memcpy(&ctx->state->open_stateid.other, &stateid->other,
357 NFS4_STATEID_OTHER_SIZE);
358 update_open_stateid(ctx->state, stateid, NULL, filep->f_mode);
359
360 nfs_file_set_open_context(filep, ctx);
361 put_nfs_open_context(ctx);
362
363 file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
364 res = filep;
365out:
366 return res;
367out_stateowner:
368 nfs4_put_state_owner(sp);
369out_ctx:
370 put_nfs_open_context(ctx);
371out_filep:
372 fput(filep);
373 goto out;
374}
375EXPORT_SYMBOL_GPL(nfs42_ssc_open);
376void nfs42_ssc_close(struct file *filep)
377{
378 struct nfs_open_context *ctx = nfs_file_open_context(filep);
379
380 ctx->state->flags = 0;
381}
382EXPORT_SYMBOL_GPL(nfs42_ssc_close);
6b7153da 383#endif /* CONFIG_NFS_V4_2 */
bea51b30 384
ce4ef7c0 385const struct file_operations nfs4_file_operations = {
3aa2d199 386 .read_iter = nfs_file_read,
edaf4369 387 .write_iter = nfs_file_write,
ce4ef7c0
BS
388 .mmap = nfs_file_mmap,
389 .open = nfs4_file_open,
5445b1fb 390 .flush = nfs4_file_flush,
ce4ef7c0 391 .release = nfs_file_release,
4ff79bc7 392 .fsync = nfs_file_fsync,
ce4ef7c0
BS
393 .lock = nfs_lock,
394 .flock = nfs_flock,
82c156f8 395 .splice_read = generic_file_splice_read,
4da54c21 396 .splice_write = iter_file_splice_write,
ce4ef7c0 397 .check_flags = nfs_check_flags,
1c994a09 398 .setlease = simple_nosetlease,
6b7153da 399#ifdef CONFIG_NFS_V4_2
2e72448b 400 .copy_file_range = nfs4_copy_file_range,
6b7153da
CH
401 .llseek = nfs4_file_llseek,
402 .fallocate = nfs42_fallocate,
2e5dfc99 403 .remap_file_range = nfs42_remap_file_range,
6b7153da
CH
404#else
405 .llseek = nfs_file_llseek,
406#endif
ce4ef7c0 407};