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 |
29 | nfs4_file_open(struct inode *inode, struct file *filp) |
30 | { |
31 | struct nfs_open_context *ctx; |
32 | struct dentry *dentry = file_dentry(file: 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(dentry: parent); |
59 | |
60 | ctx = alloc_nfs_open_context(dentry: file_dentry(file: filp), |
61 | f_mode: flags_to_mode(flags: openflags), filp); |
62 | err = PTR_ERR(ptr: ctx); |
63 | if (IS_ERR(ptr: 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(mapping: inode->i_mapping); |
71 | } |
72 | |
73 | inode = NFS_PROTO(inode: dir)->open_context(dir, ctx, openflags, &attr, NULL); |
74 | if (IS_ERR(ptr: inode)) { |
75 | err = PTR_ERR(ptr: 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 |
111 | nfs4_file_flush(struct file *file, fl_owner_t id) |
112 | { |
113 | struct inode *inode = file_inode(f: file); |
114 | errseq_t since; |
115 | |
116 | dprintk("NFS: flush(%pD2)\n" , file); |
117 | |
118 | nfs_inc_stats(inode, stat: 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(mapping: file->f_mapping); |
131 | nfs_wb_all(inode); |
132 | return filemap_check_wb_err(mapping: file->f_mapping, since); |
133 | } |
134 | |
135 | #ifdef CONFIG_NFS_V4_2 |
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(inode: file_inode(f: file_out), NFS_CAP_COPY) || |
150 | !nfs_server_capable(inode: file_inode(f: file_in), NFS_CAP_COPY)) |
151 | return -EOPNOTSUPP; |
152 | if (file_inode(f: file_in) == file_inode(f: 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(inode: file_inode(f: file_in))->rsize) |
158 | sync = true; |
159 | retry: |
160 | if (!nfs42_files_from_same_server(in: file_in, out: 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(size: sizeof(struct nfs42_copy_notify_res), |
168 | GFP_KERNEL); |
169 | if (unlikely(cn_resp == NULL)) |
170 | return -ENOMEM; |
171 | |
172 | ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp); |
173 | if (ret) { |
174 | ret = -EOPNOTSUPP; |
175 | goto out; |
176 | } |
177 | nss = &cn_resp->cnr_src; |
178 | cnrs = &cn_resp->cnr_stateid; |
179 | } |
180 | ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count, |
181 | nss, cnrs, sync); |
182 | out: |
183 | kfree(objp: cn_resp); |
184 | |
185 | if (ret == -EAGAIN) |
186 | goto retry; |
187 | return ret; |
188 | } |
189 | |
190 | static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in, |
191 | struct file *file_out, loff_t pos_out, |
192 | size_t count, unsigned int flags) |
193 | { |
194 | ssize_t ret; |
195 | |
196 | ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count, |
197 | flags); |
198 | if (ret == -EOPNOTSUPP || ret == -EXDEV) |
199 | ret = splice_copy_file_range(in: file_in, pos_in, out: file_out, |
200 | pos_out, len: count); |
201 | return ret; |
202 | } |
203 | |
204 | static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence) |
205 | { |
206 | loff_t ret; |
207 | |
208 | switch (whence) { |
209 | case SEEK_HOLE: |
210 | case SEEK_DATA: |
211 | ret = nfs42_proc_llseek(filep, offset, whence); |
212 | if (ret != -EOPNOTSUPP) |
213 | return ret; |
214 | fallthrough; |
215 | default: |
216 | return nfs_file_llseek(filep, offset, whence); |
217 | } |
218 | } |
219 | |
220 | static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len) |
221 | { |
222 | struct inode *inode = file_inode(f: filep); |
223 | long ret; |
224 | |
225 | if (!S_ISREG(inode->i_mode)) |
226 | return -EOPNOTSUPP; |
227 | |
228 | if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))) |
229 | return -EOPNOTSUPP; |
230 | |
231 | ret = inode_newsize_ok(inode, offset: offset + len); |
232 | if (ret < 0) |
233 | return ret; |
234 | |
235 | if (mode & FALLOC_FL_PUNCH_HOLE) |
236 | return nfs42_proc_deallocate(filep, offset, len); |
237 | return nfs42_proc_allocate(filep, offset, len); |
238 | } |
239 | |
240 | static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off, |
241 | struct file *dst_file, loff_t dst_off, loff_t count, |
242 | unsigned int remap_flags) |
243 | { |
244 | struct inode *dst_inode = file_inode(f: dst_file); |
245 | struct nfs_server *server = NFS_SERVER(inode: dst_inode); |
246 | struct inode *src_inode = file_inode(f: src_file); |
247 | unsigned int bs = server->clone_blksize; |
248 | bool same_inode = false; |
249 | int ret; |
250 | |
251 | /* NFS does not support deduplication. */ |
252 | if (remap_flags & REMAP_FILE_DEDUP) |
253 | return -EOPNOTSUPP; |
254 | |
255 | if (remap_flags & ~REMAP_FILE_ADVISORY) |
256 | return -EINVAL; |
257 | |
258 | if (IS_SWAPFILE(dst_inode) || IS_SWAPFILE(src_inode)) |
259 | return -ETXTBSY; |
260 | |
261 | /* check alignment w.r.t. clone_blksize */ |
262 | ret = -EINVAL; |
263 | if (bs) { |
264 | if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs)) |
265 | goto out; |
266 | if (!IS_ALIGNED(count, bs) && i_size_read(inode: src_inode) != (src_off + count)) |
267 | goto out; |
268 | } |
269 | |
270 | if (src_inode == dst_inode) |
271 | same_inode = true; |
272 | |
273 | /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */ |
274 | if (same_inode) { |
275 | inode_lock(inode: src_inode); |
276 | } else if (dst_inode < src_inode) { |
277 | inode_lock_nested(inode: dst_inode, subclass: I_MUTEX_PARENT); |
278 | inode_lock_nested(inode: src_inode, subclass: I_MUTEX_CHILD); |
279 | } else { |
280 | inode_lock_nested(inode: src_inode, subclass: I_MUTEX_PARENT); |
281 | inode_lock_nested(inode: dst_inode, subclass: I_MUTEX_CHILD); |
282 | } |
283 | |
284 | /* flush all pending writes on both src and dst so that server |
285 | * has the latest data */ |
286 | ret = nfs_sync_inode(inode: src_inode); |
287 | if (ret) |
288 | goto out_unlock; |
289 | ret = nfs_sync_inode(inode: dst_inode); |
290 | if (ret) |
291 | goto out_unlock; |
292 | |
293 | ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count); |
294 | |
295 | /* truncate inode page cache of the dst range so that future reads can fetch |
296 | * new data from server */ |
297 | if (!ret) |
298 | truncate_inode_pages_range(&dst_inode->i_data, lstart: dst_off, lend: dst_off + count - 1); |
299 | |
300 | out_unlock: |
301 | if (same_inode) { |
302 | inode_unlock(inode: src_inode); |
303 | } else if (dst_inode < src_inode) { |
304 | inode_unlock(inode: src_inode); |
305 | inode_unlock(inode: dst_inode); |
306 | } else { |
307 | inode_unlock(inode: dst_inode); |
308 | inode_unlock(inode: src_inode); |
309 | } |
310 | out: |
311 | return ret < 0 ? ret : count; |
312 | } |
313 | |
314 | static int read_name_gen = 1; |
315 | #define SSC_READ_NAME_BODY "ssc_read_%d" |
316 | |
317 | static struct file *__nfs42_ssc_open(struct vfsmount *ss_mnt, |
318 | struct nfs_fh *src_fh, nfs4_stateid *stateid) |
319 | { |
320 | struct nfs_fattr *fattr = nfs_alloc_fattr(); |
321 | struct file *filep, *res; |
322 | struct nfs_server *server; |
323 | struct inode *r_ino = NULL; |
324 | struct nfs_open_context *ctx; |
325 | struct nfs4_state_owner *sp; |
326 | char *read_name = NULL; |
327 | int len, status = 0; |
328 | |
329 | server = NFS_SB(s: ss_mnt->mnt_sb); |
330 | |
331 | if (!fattr) |
332 | return ERR_PTR(error: -ENOMEM); |
333 | |
334 | status = nfs4_proc_getattr(server, fhandle: src_fh, fattr, NULL); |
335 | if (status < 0) { |
336 | res = ERR_PTR(error: status); |
337 | goto out; |
338 | } |
339 | |
340 | if (!S_ISREG(fattr->mode)) { |
341 | res = ERR_PTR(error: -EBADF); |
342 | goto out; |
343 | } |
344 | |
345 | res = ERR_PTR(error: -ENOMEM); |
346 | len = strlen(SSC_READ_NAME_BODY) + 16; |
347 | read_name = kzalloc(size: len, GFP_KERNEL); |
348 | if (read_name == NULL) |
349 | goto out; |
350 | snprintf(buf: read_name, size: len, SSC_READ_NAME_BODY, read_name_gen++); |
351 | |
352 | r_ino = nfs_fhget(ss_mnt->mnt_sb, src_fh, fattr); |
353 | if (IS_ERR(ptr: r_ino)) { |
354 | res = ERR_CAST(ptr: r_ino); |
355 | goto out_free_name; |
356 | } |
357 | |
358 | filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, O_RDONLY, |
359 | r_ino->i_fop); |
360 | if (IS_ERR(ptr: filep)) { |
361 | res = ERR_CAST(ptr: filep); |
362 | iput(r_ino); |
363 | goto out_free_name; |
364 | } |
365 | |
366 | ctx = alloc_nfs_open_context(dentry: filep->f_path.dentry, |
367 | f_mode: flags_to_mode(flags: filep->f_flags), filp: filep); |
368 | if (IS_ERR(ptr: ctx)) { |
369 | res = ERR_CAST(ptr: ctx); |
370 | goto out_filep; |
371 | } |
372 | |
373 | res = ERR_PTR(error: -EINVAL); |
374 | sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL); |
375 | if (sp == NULL) |
376 | goto out_ctx; |
377 | |
378 | ctx->state = nfs4_get_open_state(r_ino, sp); |
379 | if (ctx->state == NULL) |
380 | goto out_stateowner; |
381 | |
382 | set_bit(nr: NFS_SRV_SSC_COPY_STATE, addr: &ctx->state->flags); |
383 | memcpy(&ctx->state->open_stateid.other, &stateid->other, |
384 | NFS4_STATEID_OTHER_SIZE); |
385 | update_open_stateid(state: ctx->state, open_stateid: stateid, NULL, fmode: filep->f_mode); |
386 | set_bit(nr: NFS_OPEN_STATE, addr: &ctx->state->flags); |
387 | |
388 | nfs_file_set_open_context(filp: filep, ctx); |
389 | put_nfs_open_context(ctx); |
390 | |
391 | file_ra_state_init(ra: &filep->f_ra, mapping: filep->f_mapping->host->i_mapping); |
392 | res = filep; |
393 | out_free_name: |
394 | kfree(objp: read_name); |
395 | out: |
396 | nfs_free_fattr(fattr); |
397 | return res; |
398 | out_stateowner: |
399 | nfs4_put_state_owner(sp); |
400 | out_ctx: |
401 | put_nfs_open_context(ctx); |
402 | out_filep: |
403 | fput(filep); |
404 | goto out_free_name; |
405 | } |
406 | |
407 | static void __nfs42_ssc_close(struct file *filep) |
408 | { |
409 | struct nfs_open_context *ctx = nfs_file_open_context(filp: filep); |
410 | |
411 | ctx->state->flags = 0; |
412 | } |
413 | |
414 | static const struct nfs4_ssc_client_ops nfs4_ssc_clnt_ops_tbl = { |
415 | .sco_open = __nfs42_ssc_open, |
416 | .sco_close = __nfs42_ssc_close, |
417 | }; |
418 | |
419 | /** |
420 | * nfs42_ssc_register_ops - Wrapper to register NFS_V4 ops in nfs_common |
421 | * |
422 | * Return values: |
423 | * None |
424 | */ |
425 | void nfs42_ssc_register_ops(void) |
426 | { |
427 | nfs42_ssc_register(ops: &nfs4_ssc_clnt_ops_tbl); |
428 | } |
429 | |
430 | /** |
431 | * nfs42_ssc_unregister_ops - wrapper to un-register NFS_V4 ops in nfs_common |
432 | * |
433 | * Return values: |
434 | * None. |
435 | */ |
436 | void nfs42_ssc_unregister_ops(void) |
437 | { |
438 | nfs42_ssc_unregister(ops: &nfs4_ssc_clnt_ops_tbl); |
439 | } |
440 | #endif /* CONFIG_NFS_V4_2 */ |
441 | |
442 | static int nfs4_setlease(struct file *file, int arg, struct file_lease **lease, |
443 | void **priv) |
444 | { |
445 | return nfs4_proc_setlease(file, arg, lease, priv); |
446 | } |
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 = nfs_file_splice_read, |
459 | .splice_write = iter_file_splice_write, |
460 | .check_flags = nfs_check_flags, |
461 | .setlease = nfs4_setlease, |
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 | |