xref: /openbmc/linux/fs/9p/vfs_file.c (revision 745b7908)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * This file contians vfs file ops for 9P2000.
4  *
5  *  Copyright (C) 2004 by Eric Van Hensbergen <ericvh@gmail.com>
6  *  Copyright (C) 2002 by Ron Minnich <rminnich@lanl.gov>
7  */
8 
9 #include <linux/module.h>
10 #include <linux/errno.h>
11 #include <linux/fs.h>
12 #include <linux/filelock.h>
13 #include <linux/sched.h>
14 #include <linux/file.h>
15 #include <linux/stat.h>
16 #include <linux/string.h>
17 #include <linux/list.h>
18 #include <linux/pagemap.h>
19 #include <linux/utsname.h>
20 #include <linux/uaccess.h>
21 #include <linux/uio.h>
22 #include <linux/slab.h>
23 #include <net/9p/9p.h>
24 #include <net/9p/client.h>
25 
26 #include "v9fs.h"
27 #include "v9fs_vfs.h"
28 #include "fid.h"
29 #include "cache.h"
30 
31 static const struct vm_operations_struct v9fs_mmap_file_vm_ops;
32 
33 /**
34  * v9fs_file_open - open a file (or directory)
35  * @inode: inode to be opened
36  * @file: file being opened
37  *
38  */
39 
40 int v9fs_file_open(struct inode *inode, struct file *file)
41 {
42 	int err;
43 	struct v9fs_session_info *v9ses;
44 	struct p9_fid *fid;
45 	int omode;
46 
47 	p9_debug(P9_DEBUG_VFS, "inode: %p file: %p\n", inode, file);
48 	v9ses = v9fs_inode2v9ses(inode);
49 	if (v9fs_proto_dotl(v9ses))
50 		omode = v9fs_open_to_dotl_flags(file->f_flags);
51 	else
52 		omode = v9fs_uflags2omode(file->f_flags,
53 					v9fs_proto_dotu(v9ses));
54 	fid = file->private_data;
55 	if (!fid) {
56 		fid = v9fs_fid_clone(file_dentry(file));
57 		if (IS_ERR(fid))
58 			return PTR_ERR(fid);
59 
60 		if ((v9ses->cache & CACHE_WRITEBACK) && (omode & P9_OWRITE)) {
61 			int writeback_omode = (omode & ~P9_OWRITE) | P9_ORDWR;
62 
63 			p9_debug(P9_DEBUG_CACHE, "write-only file with writeback enabled, try opening O_RDWR\n");
64 			err = p9_client_open(fid, writeback_omode);
65 			if (err < 0) {
66 				p9_debug(P9_DEBUG_CACHE, "could not open O_RDWR, disabling caches\n");
67 				err = p9_client_open(fid, omode);
68 				fid->mode |= P9L_DIRECT;
69 			}
70 		} else {
71 			err = p9_client_open(fid, omode);
72 		}
73 		if (err < 0) {
74 			p9_fid_put(fid);
75 			return err;
76 		}
77 		if ((file->f_flags & O_APPEND) &&
78 			(!v9fs_proto_dotu(v9ses) && !v9fs_proto_dotl(v9ses)))
79 			generic_file_llseek(file, 0, SEEK_END);
80 
81 		file->private_data = fid;
82 	}
83 
84 #ifdef CONFIG_9P_FSCACHE
85 	if (v9ses->cache & CACHE_FSCACHE)
86 		fscache_use_cookie(v9fs_inode_cookie(V9FS_I(inode)),
87 				   file->f_mode & FMODE_WRITE);
88 #endif
89 	v9fs_fid_add_modes(fid, v9ses->flags, v9ses->cache, file->f_flags);
90 	v9fs_open_fid_add(inode, &fid);
91 	return 0;
92 }
93 
94 /**
95  * v9fs_file_lock - lock a file (or directory)
96  * @filp: file to be locked
97  * @cmd: lock command
98  * @fl: file lock structure
99  *
100  * Bugs: this looks like a local only lock, we should extend into 9P
101  *       by using open exclusive
102  */
103 
104 static int v9fs_file_lock(struct file *filp, int cmd, struct file_lock *fl)
105 {
106 	struct inode *inode = file_inode(filp);
107 
108 	p9_debug(P9_DEBUG_VFS, "filp: %p lock: %p\n", filp, fl);
109 
110 	if ((IS_SETLK(cmd) || IS_SETLKW(cmd)) && fl->fl_type != F_UNLCK) {
111 		filemap_write_and_wait(inode->i_mapping);
112 		invalidate_mapping_pages(&inode->i_data, 0, -1);
113 	}
114 
115 	return 0;
116 }
117 
118 static int v9fs_file_do_lock(struct file *filp, int cmd, struct file_lock *fl)
119 {
120 	struct p9_flock flock;
121 	struct p9_fid *fid;
122 	uint8_t status = P9_LOCK_ERROR;
123 	int res = 0;
124 	unsigned char fl_type;
125 	struct v9fs_session_info *v9ses;
126 
127 	fid = filp->private_data;
128 	BUG_ON(fid == NULL);
129 
130 	BUG_ON((fl->fl_flags & FL_POSIX) != FL_POSIX);
131 
132 	res = locks_lock_file_wait(filp, fl);
133 	if (res < 0)
134 		goto out;
135 
136 	/* convert posix lock to p9 tlock args */
137 	memset(&flock, 0, sizeof(flock));
138 	/* map the lock type */
139 	switch (fl->fl_type) {
140 	case F_RDLCK:
141 		flock.type = P9_LOCK_TYPE_RDLCK;
142 		break;
143 	case F_WRLCK:
144 		flock.type = P9_LOCK_TYPE_WRLCK;
145 		break;
146 	case F_UNLCK:
147 		flock.type = P9_LOCK_TYPE_UNLCK;
148 		break;
149 	}
150 	flock.start = fl->fl_start;
151 	if (fl->fl_end == OFFSET_MAX)
152 		flock.length = 0;
153 	else
154 		flock.length = fl->fl_end - fl->fl_start + 1;
155 	flock.proc_id = fl->fl_pid;
156 	flock.client_id = fid->clnt->name;
157 	if (IS_SETLKW(cmd))
158 		flock.flags = P9_LOCK_FLAGS_BLOCK;
159 
160 	v9ses = v9fs_inode2v9ses(file_inode(filp));
161 
162 	/*
163 	 * if its a blocked request and we get P9_LOCK_BLOCKED as the status
164 	 * for lock request, keep on trying
165 	 */
166 	for (;;) {
167 		res = p9_client_lock_dotl(fid, &flock, &status);
168 		if (res < 0)
169 			goto out_unlock;
170 
171 		if (status != P9_LOCK_BLOCKED)
172 			break;
173 		if (status == P9_LOCK_BLOCKED && !IS_SETLKW(cmd))
174 			break;
175 		if (schedule_timeout_interruptible(v9ses->session_lock_timeout)
176 				!= 0)
177 			break;
178 		/*
179 		 * p9_client_lock_dotl overwrites flock.client_id with the
180 		 * server message, free and reuse the client name
181 		 */
182 		if (flock.client_id != fid->clnt->name) {
183 			kfree(flock.client_id);
184 			flock.client_id = fid->clnt->name;
185 		}
186 	}
187 
188 	/* map 9p status to VFS status */
189 	switch (status) {
190 	case P9_LOCK_SUCCESS:
191 		res = 0;
192 		break;
193 	case P9_LOCK_BLOCKED:
194 		res = -EAGAIN;
195 		break;
196 	default:
197 		WARN_ONCE(1, "unknown lock status code: %d\n", status);
198 		fallthrough;
199 	case P9_LOCK_ERROR:
200 	case P9_LOCK_GRACE:
201 		res = -ENOLCK;
202 		break;
203 	}
204 
205 out_unlock:
206 	/*
207 	 * incase server returned error for lock request, revert
208 	 * it locally
209 	 */
210 	if (res < 0 && fl->fl_type != F_UNLCK) {
211 		fl_type = fl->fl_type;
212 		fl->fl_type = F_UNLCK;
213 		/* Even if this fails we want to return the remote error */
214 		locks_lock_file_wait(filp, fl);
215 		fl->fl_type = fl_type;
216 	}
217 	if (flock.client_id != fid->clnt->name)
218 		kfree(flock.client_id);
219 out:
220 	return res;
221 }
222 
223 static int v9fs_file_getlock(struct file *filp, struct file_lock *fl)
224 {
225 	struct p9_getlock glock;
226 	struct p9_fid *fid;
227 	int res = 0;
228 
229 	fid = filp->private_data;
230 	BUG_ON(fid == NULL);
231 
232 	posix_test_lock(filp, fl);
233 	/*
234 	 * if we have a conflicting lock locally, no need to validate
235 	 * with server
236 	 */
237 	if (fl->fl_type != F_UNLCK)
238 		return res;
239 
240 	/* convert posix lock to p9 tgetlock args */
241 	memset(&glock, 0, sizeof(glock));
242 	glock.type  = P9_LOCK_TYPE_UNLCK;
243 	glock.start = fl->fl_start;
244 	if (fl->fl_end == OFFSET_MAX)
245 		glock.length = 0;
246 	else
247 		glock.length = fl->fl_end - fl->fl_start + 1;
248 	glock.proc_id = fl->fl_pid;
249 	glock.client_id = fid->clnt->name;
250 
251 	res = p9_client_getlock_dotl(fid, &glock);
252 	if (res < 0)
253 		goto out;
254 	/* map 9p lock type to os lock type */
255 	switch (glock.type) {
256 	case P9_LOCK_TYPE_RDLCK:
257 		fl->fl_type = F_RDLCK;
258 		break;
259 	case P9_LOCK_TYPE_WRLCK:
260 		fl->fl_type = F_WRLCK;
261 		break;
262 	case P9_LOCK_TYPE_UNLCK:
263 		fl->fl_type = F_UNLCK;
264 		break;
265 	}
266 	if (glock.type != P9_LOCK_TYPE_UNLCK) {
267 		fl->fl_start = glock.start;
268 		if (glock.length == 0)
269 			fl->fl_end = OFFSET_MAX;
270 		else
271 			fl->fl_end = glock.start + glock.length - 1;
272 		fl->fl_pid = -glock.proc_id;
273 	}
274 out:
275 	if (glock.client_id != fid->clnt->name)
276 		kfree(glock.client_id);
277 	return res;
278 }
279 
280 /**
281  * v9fs_file_lock_dotl - lock a file (or directory)
282  * @filp: file to be locked
283  * @cmd: lock command
284  * @fl: file lock structure
285  *
286  */
287 
288 static int v9fs_file_lock_dotl(struct file *filp, int cmd, struct file_lock *fl)
289 {
290 	struct inode *inode = file_inode(filp);
291 	int ret = -ENOLCK;
292 
293 	p9_debug(P9_DEBUG_VFS, "filp: %p cmd:%d lock: %p name: %pD\n",
294 		 filp, cmd, fl, filp);
295 
296 	if ((IS_SETLK(cmd) || IS_SETLKW(cmd)) && fl->fl_type != F_UNLCK) {
297 		filemap_write_and_wait(inode->i_mapping);
298 		invalidate_mapping_pages(&inode->i_data, 0, -1);
299 	}
300 
301 	if (IS_SETLK(cmd) || IS_SETLKW(cmd))
302 		ret = v9fs_file_do_lock(filp, cmd, fl);
303 	else if (IS_GETLK(cmd))
304 		ret = v9fs_file_getlock(filp, fl);
305 	else
306 		ret = -EINVAL;
307 	return ret;
308 }
309 
310 /**
311  * v9fs_file_flock_dotl - lock a file
312  * @filp: file to be locked
313  * @cmd: lock command
314  * @fl: file lock structure
315  *
316  */
317 
318 static int v9fs_file_flock_dotl(struct file *filp, int cmd,
319 	struct file_lock *fl)
320 {
321 	struct inode *inode = file_inode(filp);
322 	int ret = -ENOLCK;
323 
324 	p9_debug(P9_DEBUG_VFS, "filp: %p cmd:%d lock: %p name: %pD\n",
325 		 filp, cmd, fl, filp);
326 
327 	if (!(fl->fl_flags & FL_FLOCK))
328 		goto out_err;
329 
330 	if ((IS_SETLK(cmd) || IS_SETLKW(cmd)) && fl->fl_type != F_UNLCK) {
331 		filemap_write_and_wait(inode->i_mapping);
332 		invalidate_mapping_pages(&inode->i_data, 0, -1);
333 	}
334 	/* Convert flock to posix lock */
335 	fl->fl_flags |= FL_POSIX;
336 	fl->fl_flags ^= FL_FLOCK;
337 
338 	if (IS_SETLK(cmd) | IS_SETLKW(cmd))
339 		ret = v9fs_file_do_lock(filp, cmd, fl);
340 	else
341 		ret = -EINVAL;
342 out_err:
343 	return ret;
344 }
345 
346 /**
347  * v9fs_file_read_iter - read from a file
348  * @iocb: The operation parameters
349  * @to: The buffer to read into
350  *
351  */
352 static ssize_t
353 v9fs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
354 {
355 	struct p9_fid *fid = iocb->ki_filp->private_data;
356 	int ret, err = 0;
357 
358 	p9_debug(P9_DEBUG_VFS, "fid %d count %zu offset %lld\n",
359 		 fid->fid, iov_iter_count(to), iocb->ki_pos);
360 
361 	if (!(fid->mode & P9L_DIRECT)) {
362 		p9_debug(P9_DEBUG_VFS, "(cached)\n");
363 		return generic_file_read_iter(iocb, to);
364 	}
365 
366 	if (iocb->ki_filp->f_flags & O_NONBLOCK)
367 		ret = p9_client_read_once(fid, iocb->ki_pos, to, &err);
368 	else
369 		ret = p9_client_read(fid, iocb->ki_pos, to, &err);
370 	if (!ret)
371 		return err;
372 
373 	iocb->ki_pos += ret;
374 	return ret;
375 }
376 
377 /**
378  * v9fs_file_write_iter - write to a file
379  * @iocb: The operation parameters
380  * @from: The data to write
381  *
382  */
383 static ssize_t
384 v9fs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
385 {
386 	struct file *file = iocb->ki_filp;
387 	struct p9_fid *fid = file->private_data;
388 	ssize_t retval;
389 	loff_t origin;
390 	int err = 0;
391 
392 	p9_debug(P9_DEBUG_VFS, "fid %d\n", fid->fid);
393 
394 	if (!(fid->mode & (P9L_DIRECT | P9L_NOWRITECACHE))) {
395 		p9_debug(P9_DEBUG_CACHE, "(cached)\n");
396 		return generic_file_write_iter(iocb, from);
397 	}
398 
399 	retval = generic_write_checks(iocb, from);
400 	if (retval <= 0)
401 		return retval;
402 
403 	origin = iocb->ki_pos;
404 	retval = p9_client_write(file->private_data, iocb->ki_pos, from, &err);
405 	if (retval > 0) {
406 		struct inode *inode = file_inode(file);
407 		loff_t i_size;
408 		unsigned long pg_start, pg_end;
409 
410 		pg_start = origin >> PAGE_SHIFT;
411 		pg_end = (origin + retval - 1) >> PAGE_SHIFT;
412 		if (inode->i_mapping && inode->i_mapping->nrpages)
413 			invalidate_inode_pages2_range(inode->i_mapping,
414 						      pg_start, pg_end);
415 		iocb->ki_pos += retval;
416 		i_size = i_size_read(inode);
417 		if (iocb->ki_pos > i_size) {
418 			inode_add_bytes(inode, iocb->ki_pos - i_size);
419 			/*
420 			 * Need to serialize against i_size_write() in
421 			 * v9fs_stat2inode()
422 			 */
423 			v9fs_i_size_write(inode, iocb->ki_pos);
424 		}
425 		return retval;
426 	}
427 	return err;
428 }
429 
430 static int v9fs_file_fsync(struct file *filp, loff_t start, loff_t end,
431 			   int datasync)
432 {
433 	struct p9_fid *fid;
434 	struct inode *inode = filp->f_mapping->host;
435 	struct p9_wstat wstat;
436 	int retval;
437 
438 	retval = file_write_and_wait_range(filp, start, end);
439 	if (retval)
440 		return retval;
441 
442 	inode_lock(inode);
443 	p9_debug(P9_DEBUG_VFS, "filp %p datasync %x\n", filp, datasync);
444 
445 	fid = filp->private_data;
446 	v9fs_blank_wstat(&wstat);
447 
448 	retval = p9_client_wstat(fid, &wstat);
449 	inode_unlock(inode);
450 
451 	return retval;
452 }
453 
454 int v9fs_file_fsync_dotl(struct file *filp, loff_t start, loff_t end,
455 			 int datasync)
456 {
457 	struct p9_fid *fid;
458 	struct inode *inode = filp->f_mapping->host;
459 	int retval;
460 
461 	retval = file_write_and_wait_range(filp, start, end);
462 	if (retval)
463 		return retval;
464 
465 	inode_lock(inode);
466 	p9_debug(P9_DEBUG_VFS, "filp %p datasync %x\n", filp, datasync);
467 
468 	fid = filp->private_data;
469 
470 	retval = p9_client_fsync(fid, datasync);
471 	inode_unlock(inode);
472 
473 	return retval;
474 }
475 
476 static int
477 v9fs_file_mmap(struct file *filp, struct vm_area_struct *vma)
478 {
479 	int retval;
480 	struct inode *inode = file_inode(filp);
481 	struct v9fs_session_info *v9ses = v9fs_inode2v9ses(inode);
482 
483 	p9_debug(P9_DEBUG_MMAP, "filp :%p\n", filp);
484 
485 	if (!(v9ses->cache & CACHE_WRITEBACK)) {
486 		p9_debug(P9_DEBUG_CACHE, "(no mmap mode)");
487 		if (vma->vm_flags & VM_MAYSHARE)
488 			return -ENODEV;
489 		invalidate_inode_pages2(filp->f_mapping);
490 		return generic_file_readonly_mmap(filp, vma);
491 	}
492 
493 	retval = generic_file_mmap(filp, vma);
494 	if (!retval)
495 		vma->vm_ops = &v9fs_mmap_file_vm_ops;
496 
497 	return retval;
498 }
499 
500 static vm_fault_t
501 v9fs_vm_page_mkwrite(struct vm_fault *vmf)
502 {
503 	struct folio *folio = page_folio(vmf->page);
504 	struct file *filp = vmf->vma->vm_file;
505 	struct inode *inode = file_inode(filp);
506 
507 
508 	p9_debug(P9_DEBUG_VFS, "folio %p fid %lx\n",
509 		 folio, (unsigned long)filp->private_data);
510 
511 	/* Wait for the page to be written to the cache before we allow it to
512 	 * be modified.  We then assume the entire page will need writing back.
513 	 */
514 #ifdef CONFIG_9P_FSCACHE
515 	if (folio_test_fscache(folio) &&
516 	    folio_wait_fscache_killable(folio) < 0)
517 		return VM_FAULT_NOPAGE;
518 #endif
519 
520 	/* Update file times before taking page lock */
521 	file_update_time(filp);
522 
523 	if (folio_lock_killable(folio) < 0)
524 		return VM_FAULT_RETRY;
525 	if (folio_mapping(folio) != inode->i_mapping)
526 		goto out_unlock;
527 	folio_wait_stable(folio);
528 
529 	return VM_FAULT_LOCKED;
530 out_unlock:
531 	folio_unlock(folio);
532 	return VM_FAULT_NOPAGE;
533 }
534 
535 static void v9fs_mmap_vm_close(struct vm_area_struct *vma)
536 {
537 	struct inode *inode;
538 
539 	struct writeback_control wbc = {
540 		.nr_to_write = LONG_MAX,
541 		.sync_mode = WB_SYNC_ALL,
542 		.range_start = (loff_t)vma->vm_pgoff * PAGE_SIZE,
543 		 /* absolute end, byte at end included */
544 		.range_end = (loff_t)vma->vm_pgoff * PAGE_SIZE +
545 			(vma->vm_end - vma->vm_start - 1),
546 	};
547 
548 	if (!(vma->vm_flags & VM_SHARED))
549 		return;
550 
551 	p9_debug(P9_DEBUG_VFS, "9p VMA close, %p, flushing", vma);
552 
553 	inode = file_inode(vma->vm_file);
554 	filemap_fdatawrite_wbc(inode->i_mapping, &wbc);
555 }
556 
557 static const struct vm_operations_struct v9fs_mmap_file_vm_ops = {
558 	.close = v9fs_mmap_vm_close,
559 	.fault = filemap_fault,
560 	.map_pages = filemap_map_pages,
561 	.page_mkwrite = v9fs_vm_page_mkwrite,
562 };
563 
564 const struct file_operations v9fs_file_operations = {
565 	.llseek = generic_file_llseek,
566 	.read_iter = v9fs_file_read_iter,
567 	.write_iter = v9fs_file_write_iter,
568 	.open = v9fs_file_open,
569 	.release = v9fs_dir_release,
570 	.lock = v9fs_file_lock,
571 	.mmap = generic_file_readonly_mmap,
572 	.splice_read = generic_file_splice_read,
573 	.splice_write = iter_file_splice_write,
574 	.fsync = v9fs_file_fsync,
575 };
576 
577 const struct file_operations v9fs_file_operations_dotl = {
578 	.llseek = generic_file_llseek,
579 	.read_iter = v9fs_file_read_iter,
580 	.write_iter = v9fs_file_write_iter,
581 	.open = v9fs_file_open,
582 	.release = v9fs_dir_release,
583 	.lock = v9fs_file_lock_dotl,
584 	.flock = v9fs_file_flock_dotl,
585 	.mmap = v9fs_file_mmap,
586 	.splice_read = generic_file_splice_read,
587 	.splice_write = iter_file_splice_write,
588 	.fsync = v9fs_file_fsync_dotl,
589 };
590