1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/nfs/file.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 */
7 #include <linux/fs.h>
8 #include <linux/file.h>
9 #include <linux/falloc.h>
10 #include <linux/mount.h>
11 #include <linux/nfs_fs.h>
12 #include <linux/nfs_ssc.h>
13 #include "delegation.h"
14 #include "internal.h"
15 #include "iostat.h"
16 #include "fscache.h"
17 #include "pnfs.h"
18
19 #include "nfstrace.h"
20
21 #ifdef CONFIG_NFS_V4_2
22 #include "nfs42.h"
23 #endif
24
25 #define NFSDBG_FACILITY NFSDBG_FILE
26
27 static int
nfs4_file_open(struct inode * inode,struct file * filp)28 nfs4_file_open(struct inode *inode, struct file *filp)
29 {
30 struct nfs_open_context *ctx;
31 struct dentry *dentry = file_dentry(filp);
32 struct dentry *parent = NULL;
33 struct inode *dir;
34 unsigned openflags = filp->f_flags;
35 fmode_t f_mode;
36 struct iattr attr;
37 int err;
38
39 /*
40 * If no cached dentry exists or if it's negative, NFSv4 handled the
41 * opens in ->lookup() or ->create().
42 *
43 * We only get this far for a cached positive dentry. We skipped
44 * revalidation, so handle it here by dropping the dentry and returning
45 * -EOPENSTALE. The VFS will retry the lookup/create/open.
46 */
47
48 dprintk("NFS: open file(%pd2)\n", dentry);
49
50 err = nfs_check_flags(openflags);
51 if (err)
52 return err;
53
54 f_mode = filp->f_mode;
55 if ((openflags & O_ACCMODE) == 3)
56 f_mode |= flags_to_mode(openflags);
57
58 /* We can't create new files here */
59 openflags &= ~(O_CREAT|O_EXCL);
60
61 parent = dget_parent(dentry);
62 dir = d_inode(parent);
63
64 ctx = alloc_nfs_open_context(file_dentry(filp), f_mode, filp);
65 err = PTR_ERR(ctx);
66 if (IS_ERR(ctx))
67 goto out;
68
69 attr.ia_valid = ATTR_OPEN;
70 if (openflags & O_TRUNC) {
71 attr.ia_valid |= ATTR_SIZE;
72 attr.ia_size = 0;
73 filemap_write_and_wait(inode->i_mapping);
74 }
75
76 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
77 if (IS_ERR(inode)) {
78 err = PTR_ERR(inode);
79 switch (err) {
80 default:
81 goto out_put_ctx;
82 case -ENOENT:
83 case -ESTALE:
84 case -EISDIR:
85 case -ENOTDIR:
86 case -ELOOP:
87 goto out_drop;
88 }
89 }
90 if (inode != d_inode(dentry))
91 goto out_drop;
92
93 nfs_file_set_open_context(filp, ctx);
94 nfs_fscache_open_file(inode, filp);
95 err = 0;
96
97 out_put_ctx:
98 put_nfs_open_context(ctx);
99 out:
100 dput(parent);
101 return err;
102
103 out_drop:
104 d_drop(dentry);
105 err = -EOPENSTALE;
106 goto out_put_ctx;
107 }
108
109 /*
110 * Flush all dirty pages, and check for write errors.
111 */
112 static int
nfs4_file_flush(struct file * file,fl_owner_t id)113 nfs4_file_flush(struct file *file, fl_owner_t id)
114 {
115 struct inode *inode = file_inode(file);
116 errseq_t since;
117
118 dprintk("NFS: flush(%pD2)\n", file);
119
120 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
121 if ((file->f_mode & FMODE_WRITE) == 0)
122 return 0;
123
124 /*
125 * If we're holding a write delegation, then check if we're required
126 * to flush the i/o on close. If not, then just start the i/o now.
127 */
128 if (!nfs4_delegation_flush_on_close(inode))
129 return filemap_fdatawrite(file->f_mapping);
130
131 /* Flush writes to the server and return any errors */
132 since = filemap_sample_wb_err(file->f_mapping);
133 nfs_wb_all(inode);
134 return filemap_check_wb_err(file->f_mapping, since);
135 }
136
137 #ifdef CONFIG_NFS_V4_2
__nfs4_copy_file_range(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,size_t count,unsigned int flags)138 static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
139 struct file *file_out, loff_t pos_out,
140 size_t count, unsigned int flags)
141 {
142 struct nfs42_copy_notify_res *cn_resp = NULL;
143 struct nl4_server *nss = NULL;
144 nfs4_stateid *cnrs = NULL;
145 ssize_t ret;
146 bool sync = false;
147
148 /* Only offload copy if superblock is the same */
149 if (file_in->f_op != &nfs4_file_operations)
150 return -EXDEV;
151 if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY) ||
152 !nfs_server_capable(file_inode(file_in), NFS_CAP_COPY))
153 return -EOPNOTSUPP;
154 if (file_inode(file_in) == file_inode(file_out))
155 return -EOPNOTSUPP;
156 /* if the copy size if smaller than 2 RPC payloads, make it
157 * synchronous
158 */
159 if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize)
160 sync = true;
161 retry:
162 if (!nfs42_files_from_same_server(file_in, file_out)) {
163 /* for inter copy, if copy size if smaller than 12 RPC
164 * payloads, fallback to traditional copy. There are
165 * 14 RPCs during an NFSv4.x mount between source/dest
166 * servers.
167 */
168 if (sync ||
169 count <= 14 * NFS_SERVER(file_inode(file_in))->rsize)
170 return -EOPNOTSUPP;
171 cn_resp = kzalloc(sizeof(struct nfs42_copy_notify_res),
172 GFP_NOFS);
173 if (unlikely(cn_resp == NULL))
174 return -ENOMEM;
175
176 ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp);
177 if (ret) {
178 ret = -EOPNOTSUPP;
179 goto out;
180 }
181 nss = &cn_resp->cnr_src;
182 cnrs = &cn_resp->cnr_stateid;
183 }
184 ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count,
185 nss, cnrs, sync);
186 out:
187 if (!nfs42_files_from_same_server(file_in, file_out))
188 kfree(cn_resp);
189 if (ret == -EAGAIN)
190 goto retry;
191 return ret;
192 }
193
nfs4_copy_file_range(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,size_t count,unsigned int flags)194 static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
195 struct file *file_out, loff_t pos_out,
196 size_t count, unsigned int flags)
197 {
198 ssize_t ret;
199
200 ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
201 flags);
202 if (ret == -EOPNOTSUPP || ret == -EXDEV)
203 ret = generic_copy_file_range(file_in, pos_in, file_out,
204 pos_out, count, flags);
205 return ret;
206 }
207
nfs4_file_llseek(struct file * filep,loff_t offset,int whence)208 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
209 {
210 loff_t ret;
211
212 switch (whence) {
213 case SEEK_HOLE:
214 case SEEK_DATA:
215 ret = nfs42_proc_llseek(filep, offset, whence);
216 if (ret != -EOPNOTSUPP)
217 return ret;
218 fallthrough;
219 default:
220 return nfs_file_llseek(filep, offset, whence);
221 }
222 }
223
nfs42_fallocate(struct file * filep,int mode,loff_t offset,loff_t len)224 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
225 {
226 struct inode *inode = file_inode(filep);
227 long ret;
228
229 if (!S_ISREG(inode->i_mode))
230 return -EOPNOTSUPP;
231
232 if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
233 return -EOPNOTSUPP;
234
235 ret = inode_newsize_ok(inode, offset + len);
236 if (ret < 0)
237 return ret;
238
239 if (mode & FALLOC_FL_PUNCH_HOLE)
240 return nfs42_proc_deallocate(filep, offset, len);
241 return nfs42_proc_allocate(filep, offset, len);
242 }
243
nfs42_remap_file_range(struct file * src_file,loff_t src_off,struct file * dst_file,loff_t dst_off,loff_t count,unsigned int remap_flags)244 static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
245 struct file *dst_file, loff_t dst_off, loff_t count,
246 unsigned int remap_flags)
247 {
248 struct inode *dst_inode = file_inode(dst_file);
249 struct nfs_server *server = NFS_SERVER(dst_inode);
250 struct inode *src_inode = file_inode(src_file);
251 unsigned int bs = server->clone_blksize;
252 bool same_inode = false;
253 int ret;
254
255 /* NFS does not support deduplication. */
256 if (remap_flags & REMAP_FILE_DEDUP)
257 return -EOPNOTSUPP;
258
259 if (remap_flags & ~REMAP_FILE_ADVISORY)
260 return -EINVAL;
261
262 if (IS_SWAPFILE(dst_inode) || IS_SWAPFILE(src_inode))
263 return -ETXTBSY;
264
265 /* check alignment w.r.t. clone_blksize */
266 ret = -EINVAL;
267 if (bs) {
268 if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
269 goto out;
270 if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
271 goto out;
272 }
273
274 if (src_inode == dst_inode)
275 same_inode = true;
276
277 /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
278 if (same_inode) {
279 inode_lock(src_inode);
280 } else if (dst_inode < src_inode) {
281 inode_lock_nested(dst_inode, I_MUTEX_PARENT);
282 inode_lock_nested(src_inode, I_MUTEX_CHILD);
283 } else {
284 inode_lock_nested(src_inode, I_MUTEX_PARENT);
285 inode_lock_nested(dst_inode, I_MUTEX_CHILD);
286 }
287
288 /* flush all pending writes on both src and dst so that server
289 * has the latest data */
290 ret = nfs_sync_inode(src_inode);
291 if (ret)
292 goto out_unlock;
293 ret = nfs_sync_inode(dst_inode);
294 if (ret)
295 goto out_unlock;
296
297 ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
298
299 /* truncate inode page cache of the dst range so that future reads can fetch
300 * new data from server */
301 if (!ret)
302 truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
303
304 out_unlock:
305 if (same_inode) {
306 inode_unlock(src_inode);
307 } else if (dst_inode < src_inode) {
308 inode_unlock(src_inode);
309 inode_unlock(dst_inode);
310 } else {
311 inode_unlock(dst_inode);
312 inode_unlock(src_inode);
313 }
314 out:
315 return ret < 0 ? ret : count;
316 }
317
318 static int read_name_gen = 1;
319 #define SSC_READ_NAME_BODY "ssc_read_%d"
320
__nfs42_ssc_open(struct vfsmount * ss_mnt,struct nfs_fh * src_fh,nfs4_stateid * stateid)321 static struct file *__nfs42_ssc_open(struct vfsmount *ss_mnt,
322 struct nfs_fh *src_fh, nfs4_stateid *stateid)
323 {
324 struct nfs_fattr *fattr = nfs_alloc_fattr();
325 struct file *filep, *res;
326 struct nfs_server *server;
327 struct inode *r_ino = NULL;
328 struct nfs_open_context *ctx;
329 struct nfs4_state_owner *sp;
330 char *read_name = NULL;
331 int len, status = 0;
332
333 server = NFS_SERVER(ss_mnt->mnt_root->d_inode);
334
335 if (!fattr)
336 return ERR_PTR(-ENOMEM);
337
338 status = nfs4_proc_getattr(server, src_fh, fattr, NULL, NULL);
339 if (status < 0) {
340 res = ERR_PTR(status);
341 goto out;
342 }
343
344 if (!S_ISREG(fattr->mode)) {
345 res = ERR_PTR(-EBADF);
346 goto out;
347 }
348
349 res = ERR_PTR(-ENOMEM);
350 len = strlen(SSC_READ_NAME_BODY) + 16;
351 read_name = kzalloc(len, GFP_NOFS);
352 if (read_name == NULL)
353 goto out;
354 snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
355
356 r_ino = nfs_fhget(ss_mnt->mnt_root->d_inode->i_sb, src_fh, fattr,
357 NULL);
358 if (IS_ERR(r_ino)) {
359 res = ERR_CAST(r_ino);
360 goto out_free_name;
361 }
362
363 filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, FMODE_READ,
364 r_ino->i_fop);
365 if (IS_ERR(filep)) {
366 res = ERR_CAST(filep);
367 iput(r_ino);
368 goto out_free_name;
369 }
370 filep->f_mode |= FMODE_READ;
371
372 ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode,
373 filep);
374 if (IS_ERR(ctx)) {
375 res = ERR_CAST(ctx);
376 goto out_filep;
377 }
378
379 res = ERR_PTR(-EINVAL);
380 sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
381 if (sp == NULL)
382 goto out_ctx;
383
384 ctx->state = nfs4_get_open_state(r_ino, sp);
385 if (ctx->state == NULL)
386 goto out_stateowner;
387
388 set_bit(NFS_SRV_SSC_COPY_STATE, &ctx->state->flags);
389 memcpy(&ctx->state->open_stateid.other, &stateid->other,
390 NFS4_STATEID_OTHER_SIZE);
391 update_open_stateid(ctx->state, stateid, NULL, filep->f_mode);
392 set_bit(NFS_OPEN_STATE, &ctx->state->flags);
393
394 nfs_file_set_open_context(filep, ctx);
395 put_nfs_open_context(ctx);
396
397 file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
398 res = filep;
399 out_free_name:
400 kfree(read_name);
401 out:
402 nfs_free_fattr(fattr);
403 return res;
404 out_stateowner:
405 nfs4_put_state_owner(sp);
406 out_ctx:
407 put_nfs_open_context(ctx);
408 out_filep:
409 fput(filep);
410 goto out_free_name;
411 }
412
__nfs42_ssc_close(struct file * filep)413 static void __nfs42_ssc_close(struct file *filep)
414 {
415 struct nfs_open_context *ctx = nfs_file_open_context(filep);
416
417 ctx->state->flags = 0;
418 }
419
420 static const struct nfs4_ssc_client_ops nfs4_ssc_clnt_ops_tbl = {
421 .sco_open = __nfs42_ssc_open,
422 .sco_close = __nfs42_ssc_close,
423 };
424
425 /**
426 * nfs42_ssc_register_ops - Wrapper to register NFS_V4 ops in nfs_common
427 *
428 * Return values:
429 * None
430 */
nfs42_ssc_register_ops(void)431 void nfs42_ssc_register_ops(void)
432 {
433 nfs42_ssc_register(&nfs4_ssc_clnt_ops_tbl);
434 }
435
436 /**
437 * nfs42_ssc_unregister_ops - wrapper to un-register NFS_V4 ops in nfs_common
438 *
439 * Return values:
440 * None.
441 */
nfs42_ssc_unregister_ops(void)442 void nfs42_ssc_unregister_ops(void)
443 {
444 nfs42_ssc_unregister(&nfs4_ssc_clnt_ops_tbl);
445 }
446 #endif /* CONFIG_NFS_V4_2 */
447
448 const struct file_operations nfs4_file_operations = {
449 .read_iter = nfs_file_read,
450 .write_iter = nfs_file_write,
451 .mmap = nfs_file_mmap,
452 .open = nfs4_file_open,
453 .flush = nfs4_file_flush,
454 .release = nfs_file_release,
455 .fsync = nfs_file_fsync,
456 .lock = nfs_lock,
457 .flock = nfs_flock,
458 .splice_read = generic_file_splice_read,
459 .splice_write = iter_file_splice_write,
460 .check_flags = nfs_check_flags,
461 .setlease = simple_nosetlease,
462 #ifdef CONFIG_NFS_V4_2
463 .copy_file_range = nfs4_copy_file_range,
464 .llseek = nfs4_file_llseek,
465 .fallocate = nfs42_fallocate,
466 .remap_file_range = nfs42_remap_file_range,
467 #else
468 .llseek = nfs_file_llseek,
469 #endif
470 };
471