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 <linux/splice.h>
14 #include "delegation.h"
15 #include "internal.h"
16 #include "iostat.h"
17 #include "fscache.h"
18 #include "pnfs.h"
19
20 #include "nfstrace.h"
21
22 #ifdef CONFIG_NFS_V4_2
23 #include "nfs42.h"
24 #endif
25
26 #define NFSDBG_FACILITY NFSDBG_FILE
27
28 static int
nfs4_file_open(struct inode * inode,struct file * filp)29 nfs4_file_open(struct inode *inode, struct file *filp)
30 {
31 struct nfs_open_context *ctx;
32 struct dentry *dentry = file_dentry(filp);
33 struct dentry *parent = NULL;
34 struct inode *dir;
35 unsigned openflags = filp->f_flags;
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 /* We can't create new files here */
55 openflags &= ~(O_CREAT|O_EXCL);
56
57 parent = dget_parent(dentry);
58 dir = d_inode(parent);
59
60 ctx = alloc_nfs_open_context(file_dentry(filp),
61 flags_to_mode(openflags), filp);
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;
70 filemap_write_and_wait(inode->i_mapping);
71 }
72
73 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
74 if (IS_ERR(inode)) {
75 err = PTR_ERR(inode);
76 switch (err) {
77 default:
78 goto out_put_ctx;
79 case -ENOENT:
80 case -ESTALE:
81 case -EISDIR:
82 case -ENOTDIR:
83 case -ELOOP:
84 goto out_drop;
85 }
86 }
87 if (inode != d_inode(dentry))
88 goto out_drop;
89
90 nfs_file_set_open_context(filp, ctx);
91 nfs_fscache_open_file(inode, filp);
92 err = 0;
93 filp->f_mode |= FMODE_CAN_ODIRECT;
94
95 out_put_ctx:
96 put_nfs_open_context(ctx);
97 out:
98 dput(parent);
99 return err;
100
101 out_drop:
102 d_drop(dentry);
103 err = -EOPENSTALE;
104 goto out_put_ctx;
105 }
106
107 /*
108 * Flush all dirty pages, and check for write errors.
109 */
110 static int
nfs4_file_flush(struct file * file,fl_owner_t id)111 nfs4_file_flush(struct file *file, fl_owner_t id)
112 {
113 struct inode *inode = file_inode(file);
114 errseq_t since;
115
116 dprintk("NFS: flush(%pD2)\n", file);
117
118 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
119 if ((file->f_mode & FMODE_WRITE) == 0)
120 return 0;
121
122 /*
123 * If we're holding a write delegation, then check if we're required
124 * to flush the i/o on close. If not, then just start the i/o now.
125 */
126 if (!nfs4_delegation_flush_on_close(inode))
127 return filemap_fdatawrite(file->f_mapping);
128
129 /* Flush writes to the server and return any errors */
130 since = filemap_sample_wb_err(file->f_mapping);
131 nfs_wb_all(inode);
132 return filemap_check_wb_err(file->f_mapping, since);
133 }
134
135 #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)136 static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
137 struct file *file_out, loff_t pos_out,
138 size_t count, unsigned int flags)
139 {
140 struct nfs42_copy_notify_res *cn_resp = NULL;
141 struct nl4_server *nss = NULL;
142 nfs4_stateid *cnrs = NULL;
143 ssize_t ret;
144 bool sync = false;
145
146 /* Only offload copy if superblock is the same */
147 if (file_in->f_op != &nfs4_file_operations)
148 return -EXDEV;
149 if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY) ||
150 !nfs_server_capable(file_inode(file_in), NFS_CAP_COPY))
151 return -EOPNOTSUPP;
152 if (file_inode(file_in) == file_inode(file_out))
153 return -EOPNOTSUPP;
154 /* if the copy size if smaller than 2 RPC payloads, make it
155 * synchronous
156 */
157 if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize)
158 sync = true;
159 retry:
160 if (!nfs42_files_from_same_server(file_in, file_out)) {
161 /*
162 * for inter copy, if copy size is too small
163 * then fallback to generic copy.
164 */
165 if (sync)
166 return -EOPNOTSUPP;
167 cn_resp = kzalloc_obj(struct nfs42_copy_notify_res);
168 if (unlikely(cn_resp == NULL))
169 return -ENOMEM;
170
171 ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp);
172 if (ret) {
173 ret = -EOPNOTSUPP;
174 goto out;
175 }
176 nss = &cn_resp->cnr_src;
177 cnrs = &cn_resp->cnr_stateid;
178 }
179 ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count,
180 nss, cnrs, sync);
181 out:
182 kfree(cn_resp);
183
184 if (ret == -EAGAIN)
185 goto retry;
186 return ret;
187 }
188
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)189 static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
190 struct file *file_out, loff_t pos_out,
191 size_t count, unsigned int flags)
192 {
193 ssize_t ret;
194
195 ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
196 flags);
197 if (ret == -EOPNOTSUPP || ret == -EXDEV)
198 ret = splice_copy_file_range(file_in, pos_in, file_out,
199 pos_out, count);
200 return ret;
201 }
202
nfs4_file_llseek(struct file * filep,loff_t offset,int whence)203 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
204 {
205 loff_t ret;
206
207 switch (whence) {
208 case SEEK_HOLE:
209 case SEEK_DATA:
210 ret = nfs42_proc_llseek(filep, offset, whence);
211 if (ret != -EOPNOTSUPP)
212 return ret;
213 fallthrough;
214 default:
215 return nfs_file_llseek(filep, offset, whence);
216 }
217 }
218
nfs42_fallocate(struct file * filep,int mode,loff_t offset,loff_t len)219 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
220 {
221 struct inode *inode = file_inode(filep);
222 long ret;
223
224 if (!S_ISREG(inode->i_mode))
225 return -EOPNOTSUPP;
226
227 switch (mode) {
228 case 0:
229 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
230 case FALLOC_FL_ZERO_RANGE:
231 break;
232 default:
233 return -EOPNOTSUPP;
234 }
235
236 ret = inode_newsize_ok(inode, offset + len);
237 if (ret < 0)
238 return ret;
239
240 if (mode & FALLOC_FL_PUNCH_HOLE)
241 return nfs42_proc_deallocate(filep, offset, len);
242 else if (mode & FALLOC_FL_ZERO_RANGE)
243 return nfs42_proc_zero_range(filep, offset ,len);
244 return nfs42_proc_allocate(filep, offset, len);
245 }
246
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)247 static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
248 struct file *dst_file, loff_t dst_off, loff_t count,
249 unsigned int remap_flags)
250 {
251 struct inode *dst_inode = file_inode(dst_file);
252 struct nfs_server *server = NFS_SERVER(dst_inode);
253 struct inode *src_inode = file_inode(src_file);
254 unsigned int bs = server->clone_blksize;
255 int ret;
256
257 /* NFS does not support deduplication. */
258 if (remap_flags & REMAP_FILE_DEDUP)
259 return -EOPNOTSUPP;
260
261 if (remap_flags & ~REMAP_FILE_ADVISORY)
262 return -EINVAL;
263
264 if (IS_SWAPFILE(dst_inode) || IS_SWAPFILE(src_inode))
265 return -ETXTBSY;
266
267 /* check alignment w.r.t. clone_blksize */
268 ret = -EINVAL;
269 if (bs) {
270 if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
271 goto out;
272 if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
273 goto out;
274 }
275
276 /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
277 lock_two_nondirectories(src_inode, dst_inode);
278 /* flush all pending writes on both src and dst so that server
279 * has the latest data */
280 nfs_file_block_o_direct(NFS_I(src_inode));
281 ret = nfs_sync_inode(src_inode);
282 if (ret)
283 goto out_unlock;
284 nfs_file_block_o_direct(NFS_I(dst_inode));
285 ret = nfs_sync_inode(dst_inode);
286 if (ret)
287 goto out_unlock;
288
289 ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
290
291 /* truncate inode page cache of the dst range so that future reads can fetch
292 * new data from server */
293 if (!ret)
294 truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
295
296 out_unlock:
297 unlock_two_nondirectories(src_inode, dst_inode);
298 out:
299 return ret < 0 ? ret : count;
300 }
301
302 static int read_name_gen = 1;
303 #define SSC_READ_NAME_BODY "ssc_read_%d"
304
__nfs42_ssc_open(struct vfsmount * ss_mnt,struct nfs_fh * src_fh,nfs4_stateid * stateid)305 static struct file *__nfs42_ssc_open(struct vfsmount *ss_mnt,
306 struct nfs_fh *src_fh, nfs4_stateid *stateid)
307 {
308 struct nfs_fattr *fattr = nfs_alloc_fattr();
309 struct file *filep, *res;
310 struct nfs_server *server;
311 struct inode *r_ino = NULL;
312 struct nfs_open_context *ctx;
313 struct nfs4_state_owner *sp;
314 char *read_name = NULL;
315 int len, status = 0;
316
317 server = NFS_SB(ss_mnt->mnt_sb);
318
319 if (!fattr)
320 return ERR_PTR(-ENOMEM);
321
322 status = nfs4_proc_getattr(server, src_fh, fattr, NULL);
323 if (status < 0) {
324 res = ERR_PTR(status);
325 goto out;
326 }
327
328 if (!S_ISREG(fattr->mode)) {
329 res = ERR_PTR(-EBADF);
330 goto out;
331 }
332
333 res = ERR_PTR(-ENOMEM);
334 len = strlen(SSC_READ_NAME_BODY) + 16;
335 read_name = kzalloc(len, GFP_KERNEL);
336 if (read_name == NULL)
337 goto out;
338 snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
339
340 r_ino = nfs_fhget(ss_mnt->mnt_sb, src_fh, fattr);
341 if (IS_ERR(r_ino)) {
342 res = ERR_CAST(r_ino);
343 goto out_free_name;
344 }
345
346 filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, O_RDONLY,
347 r_ino->i_fop);
348 if (IS_ERR(filep)) {
349 res = ERR_CAST(filep);
350 iput(r_ino);
351 goto out_free_name;
352 }
353
354 ctx = alloc_nfs_open_context(filep->f_path.dentry,
355 flags_to_mode(filep->f_flags), filep);
356 if (IS_ERR(ctx)) {
357 res = ERR_CAST(ctx);
358 goto out_filep;
359 }
360
361 res = ERR_PTR(-EINVAL);
362 sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
363 if (sp == NULL)
364 goto out_ctx;
365
366 ctx->state = nfs4_get_open_state(r_ino, sp);
367 if (ctx->state == NULL)
368 goto out_stateowner;
369
370 set_bit(NFS_SRV_SSC_COPY_STATE, &ctx->state->flags);
371 memcpy(&ctx->state->open_stateid.other, &stateid->other,
372 NFS4_STATEID_OTHER_SIZE);
373 update_open_stateid(ctx->state, stateid, NULL, filep->f_mode);
374 set_bit(NFS_OPEN_STATE, &ctx->state->flags);
375
376 nfs_file_set_open_context(filep, ctx);
377 put_nfs_open_context(ctx);
378
379 file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
380 res = filep;
381 out_free_name:
382 kfree(read_name);
383 out:
384 nfs_free_fattr(fattr);
385 return res;
386 out_stateowner:
387 nfs4_put_state_owner(sp);
388 out_ctx:
389 put_nfs_open_context(ctx);
390 out_filep:
391 fput(filep);
392 goto out_free_name;
393 }
394
__nfs42_ssc_close(struct file * filep)395 static void __nfs42_ssc_close(struct file *filep)
396 {
397 struct nfs_open_context *ctx = nfs_file_open_context(filep);
398
399 ctx->state->flags = 0;
400 }
401
402 static const struct nfs4_ssc_client_ops nfs4_ssc_clnt_ops_tbl = {
403 .sco_open = __nfs42_ssc_open,
404 .sco_close = __nfs42_ssc_close,
405 };
406
407 /**
408 * nfs42_ssc_register_ops - Wrapper to register NFS_V4 ops in nfs_common
409 *
410 * Return values:
411 * None
412 */
nfs42_ssc_register_ops(void)413 void nfs42_ssc_register_ops(void)
414 {
415 nfs42_ssc_register(&nfs4_ssc_clnt_ops_tbl);
416 }
417
418 /**
419 * nfs42_ssc_unregister_ops - wrapper to un-register NFS_V4 ops in nfs_common
420 *
421 * Return values:
422 * None.
423 */
nfs42_ssc_unregister_ops(void)424 void nfs42_ssc_unregister_ops(void)
425 {
426 nfs42_ssc_unregister(&nfs4_ssc_clnt_ops_tbl);
427 }
428 #endif /* CONFIG_NFS_V4_2 */
429
nfs4_setlease(struct file * file,int arg,struct file_lease ** lease,void ** priv)430 static int nfs4_setlease(struct file *file, int arg, struct file_lease **lease,
431 void **priv)
432 {
433 return nfs4_proc_setlease(file, arg, lease, priv);
434 }
435
436 const struct file_operations nfs4_file_operations = {
437 .read_iter = nfs_file_read,
438 .write_iter = nfs_file_write,
439 .mmap_prepare = nfs_file_mmap_prepare,
440 .open = nfs4_file_open,
441 .flush = nfs4_file_flush,
442 .release = nfs_file_release,
443 .fsync = nfs_file_fsync,
444 .lock = nfs_lock,
445 .flock = nfs_flock,
446 .splice_read = nfs_file_splice_read,
447 .splice_write = iter_file_splice_write,
448 .check_flags = nfs_check_flags,
449 .setlease = nfs4_setlease,
450 #ifdef CONFIG_NFS_V4_2
451 .copy_file_range = nfs4_copy_file_range,
452 .llseek = nfs4_file_llseek,
453 .fallocate = nfs42_fallocate,
454 .remap_file_range = nfs42_remap_file_range,
455 #else
456 .llseek = nfs_file_llseek,
457 #endif
458 .fop_flags = FOP_DONTCACHE,
459 };
460