1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS filesystem file handling
3  *
4  * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/fs.h>
12 #include <linux/pagemap.h>
13 #include <linux/writeback.h>
14 #include <linux/gfp.h>
15 #include <linux/task_io_accounting_ops.h>
16 #include <linux/mm.h>
17 #include <linux/swap.h>
18 #include <linux/netfs.h>
19 #include "internal.h"
20 
21 static int afs_file_mmap(struct file *file, struct vm_area_struct *vma);
22 static int afs_symlink_read_folio(struct file *file, struct folio *folio);
23 static void afs_invalidate_folio(struct folio *folio, size_t offset,
24 			       size_t length);
25 static bool afs_release_folio(struct folio *folio, gfp_t gfp_flags);
26 
27 static ssize_t afs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter);
28 static void afs_vm_open(struct vm_area_struct *area);
29 static void afs_vm_close(struct vm_area_struct *area);
30 static vm_fault_t afs_vm_map_pages(struct vm_fault *vmf, pgoff_t start_pgoff, pgoff_t end_pgoff);
31 
32 const struct file_operations afs_file_operations = {
33 	.open		= afs_open,
34 	.release	= afs_release,
35 	.llseek		= generic_file_llseek,
36 	.read_iter	= afs_file_read_iter,
37 	.write_iter	= afs_file_write,
38 	.mmap		= afs_file_mmap,
39 	.splice_read	= generic_file_splice_read,
40 	.splice_write	= iter_file_splice_write,
41 	.fsync		= afs_fsync,
42 	.lock		= afs_lock,
43 	.flock		= afs_flock,
44 };
45 
46 const struct inode_operations afs_file_inode_operations = {
47 	.getattr	= afs_getattr,
48 	.setattr	= afs_setattr,
49 	.permission	= afs_permission,
50 };
51 
52 const struct address_space_operations afs_file_aops = {
53 	.read_folio	= netfs_read_folio,
54 	.readahead	= netfs_readahead,
55 	.dirty_folio	= afs_dirty_folio,
56 	.launder_folio	= afs_launder_folio,
57 	.release_folio	= afs_release_folio,
58 	.invalidate_folio = afs_invalidate_folio,
59 	.write_begin	= afs_write_begin,
60 	.write_end	= afs_write_end,
61 	.writepage	= afs_writepage,
62 	.writepages	= afs_writepages,
63 };
64 
65 const struct address_space_operations afs_symlink_aops = {
66 	.read_folio	= afs_symlink_read_folio,
67 	.release_folio	= afs_release_folio,
68 	.invalidate_folio = afs_invalidate_folio,
69 };
70 
71 static const struct vm_operations_struct afs_vm_ops = {
72 	.open		= afs_vm_open,
73 	.close		= afs_vm_close,
74 	.fault		= filemap_fault,
75 	.map_pages	= afs_vm_map_pages,
76 	.page_mkwrite	= afs_page_mkwrite,
77 };
78 
79 /*
80  * Discard a pin on a writeback key.
81  */
afs_put_wb_key(struct afs_wb_key * wbk)82 void afs_put_wb_key(struct afs_wb_key *wbk)
83 {
84 	if (wbk && refcount_dec_and_test(&wbk->usage)) {
85 		key_put(wbk->key);
86 		kfree(wbk);
87 	}
88 }
89 
90 /*
91  * Cache key for writeback.
92  */
afs_cache_wb_key(struct afs_vnode * vnode,struct afs_file * af)93 int afs_cache_wb_key(struct afs_vnode *vnode, struct afs_file *af)
94 {
95 	struct afs_wb_key *wbk, *p;
96 
97 	wbk = kzalloc(sizeof(struct afs_wb_key), GFP_KERNEL);
98 	if (!wbk)
99 		return -ENOMEM;
100 	refcount_set(&wbk->usage, 2);
101 	wbk->key = af->key;
102 
103 	spin_lock(&vnode->wb_lock);
104 	list_for_each_entry(p, &vnode->wb_keys, vnode_link) {
105 		if (p->key == wbk->key)
106 			goto found;
107 	}
108 
109 	key_get(wbk->key);
110 	list_add_tail(&wbk->vnode_link, &vnode->wb_keys);
111 	spin_unlock(&vnode->wb_lock);
112 	af->wb = wbk;
113 	return 0;
114 
115 found:
116 	refcount_inc(&p->usage);
117 	spin_unlock(&vnode->wb_lock);
118 	af->wb = p;
119 	kfree(wbk);
120 	return 0;
121 }
122 
123 /*
124  * open an AFS file or directory and attach a key to it
125  */
afs_open(struct inode * inode,struct file * file)126 int afs_open(struct inode *inode, struct file *file)
127 {
128 	struct afs_vnode *vnode = AFS_FS_I(inode);
129 	struct afs_file *af;
130 	struct key *key;
131 	int ret;
132 
133 	_enter("{%llx:%llu},", vnode->fid.vid, vnode->fid.vnode);
134 
135 	key = afs_request_key(vnode->volume->cell);
136 	if (IS_ERR(key)) {
137 		ret = PTR_ERR(key);
138 		goto error;
139 	}
140 
141 	af = kzalloc(sizeof(*af), GFP_KERNEL);
142 	if (!af) {
143 		ret = -ENOMEM;
144 		goto error_key;
145 	}
146 	af->key = key;
147 
148 	ret = afs_validate(vnode, key);
149 	if (ret < 0)
150 		goto error_af;
151 
152 	if (file->f_mode & FMODE_WRITE) {
153 		ret = afs_cache_wb_key(vnode, af);
154 		if (ret < 0)
155 			goto error_af;
156 	}
157 
158 	if (file->f_flags & O_TRUNC)
159 		set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
160 
161 	fscache_use_cookie(afs_vnode_cache(vnode), file->f_mode & FMODE_WRITE);
162 
163 	file->private_data = af;
164 	_leave(" = 0");
165 	return 0;
166 
167 error_af:
168 	kfree(af);
169 error_key:
170 	key_put(key);
171 error:
172 	_leave(" = %d", ret);
173 	return ret;
174 }
175 
176 /*
177  * release an AFS file or directory and discard its key
178  */
afs_release(struct inode * inode,struct file * file)179 int afs_release(struct inode *inode, struct file *file)
180 {
181 	struct afs_vnode_cache_aux aux;
182 	struct afs_vnode *vnode = AFS_FS_I(inode);
183 	struct afs_file *af = file->private_data;
184 	loff_t i_size;
185 	int ret = 0;
186 
187 	_enter("{%llx:%llu},", vnode->fid.vid, vnode->fid.vnode);
188 
189 	if ((file->f_mode & FMODE_WRITE))
190 		ret = vfs_fsync(file, 0);
191 
192 	file->private_data = NULL;
193 	if (af->wb)
194 		afs_put_wb_key(af->wb);
195 
196 	if ((file->f_mode & FMODE_WRITE)) {
197 		i_size = i_size_read(&vnode->netfs.inode);
198 		afs_set_cache_aux(vnode, &aux);
199 		fscache_unuse_cookie(afs_vnode_cache(vnode), &aux, &i_size);
200 	} else {
201 		fscache_unuse_cookie(afs_vnode_cache(vnode), NULL, NULL);
202 	}
203 
204 	key_put(af->key);
205 	kfree(af);
206 	afs_prune_wb_keys(vnode);
207 	_leave(" = %d", ret);
208 	return ret;
209 }
210 
211 /*
212  * Allocate a new read record.
213  */
afs_alloc_read(gfp_t gfp)214 struct afs_read *afs_alloc_read(gfp_t gfp)
215 {
216 	struct afs_read *req;
217 
218 	req = kzalloc(sizeof(struct afs_read), gfp);
219 	if (req)
220 		refcount_set(&req->usage, 1);
221 
222 	return req;
223 }
224 
225 /*
226  * Dispose of a ref to a read record.
227  */
afs_put_read(struct afs_read * req)228 void afs_put_read(struct afs_read *req)
229 {
230 	if (refcount_dec_and_test(&req->usage)) {
231 		if (req->cleanup)
232 			req->cleanup(req);
233 		key_put(req->key);
234 		kfree(req);
235 	}
236 }
237 
afs_fetch_data_notify(struct afs_operation * op)238 static void afs_fetch_data_notify(struct afs_operation *op)
239 {
240 	struct afs_read *req = op->fetch.req;
241 	struct netfs_io_subrequest *subreq = req->subreq;
242 	int error = op->error;
243 
244 	if (error == -ECONNABORTED)
245 		error = afs_abort_to_error(op->ac.abort_code);
246 	req->error = error;
247 
248 	if (subreq) {
249 		__set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
250 		netfs_subreq_terminated(subreq, error ?: req->actual_len, false);
251 		req->subreq = NULL;
252 	} else if (req->done) {
253 		req->done(req);
254 	}
255 }
256 
afs_fetch_data_success(struct afs_operation * op)257 static void afs_fetch_data_success(struct afs_operation *op)
258 {
259 	struct afs_vnode *vnode = op->file[0].vnode;
260 
261 	_enter("op=%08x", op->debug_id);
262 	afs_vnode_commit_status(op, &op->file[0]);
263 	afs_stat_v(vnode, n_fetches);
264 	atomic_long_add(op->fetch.req->actual_len, &op->net->n_fetch_bytes);
265 	afs_fetch_data_notify(op);
266 }
267 
afs_fetch_data_put(struct afs_operation * op)268 static void afs_fetch_data_put(struct afs_operation *op)
269 {
270 	op->fetch.req->error = op->error;
271 	afs_put_read(op->fetch.req);
272 }
273 
274 static const struct afs_operation_ops afs_fetch_data_operation = {
275 	.issue_afs_rpc	= afs_fs_fetch_data,
276 	.issue_yfs_rpc	= yfs_fs_fetch_data,
277 	.success	= afs_fetch_data_success,
278 	.aborted	= afs_check_for_remote_deletion,
279 	.failed		= afs_fetch_data_notify,
280 	.put		= afs_fetch_data_put,
281 };
282 
283 /*
284  * Fetch file data from the volume.
285  */
afs_fetch_data(struct afs_vnode * vnode,struct afs_read * req)286 int afs_fetch_data(struct afs_vnode *vnode, struct afs_read *req)
287 {
288 	struct afs_operation *op;
289 
290 	_enter("%s{%llx:%llu.%u},%x,,,",
291 	       vnode->volume->name,
292 	       vnode->fid.vid,
293 	       vnode->fid.vnode,
294 	       vnode->fid.unique,
295 	       key_serial(req->key));
296 
297 	op = afs_alloc_operation(req->key, vnode->volume);
298 	if (IS_ERR(op)) {
299 		if (req->subreq)
300 			netfs_subreq_terminated(req->subreq, PTR_ERR(op), false);
301 		return PTR_ERR(op);
302 	}
303 
304 	afs_op_set_vnode(op, 0, vnode);
305 
306 	op->fetch.req	= afs_get_read(req);
307 	op->ops		= &afs_fetch_data_operation;
308 	return afs_do_sync_operation(op);
309 }
310 
afs_issue_read(struct netfs_io_subrequest * subreq)311 static void afs_issue_read(struct netfs_io_subrequest *subreq)
312 {
313 	struct afs_vnode *vnode = AFS_FS_I(subreq->rreq->inode);
314 	struct afs_read *fsreq;
315 
316 	fsreq = afs_alloc_read(GFP_NOFS);
317 	if (!fsreq)
318 		return netfs_subreq_terminated(subreq, -ENOMEM, false);
319 
320 	fsreq->subreq	= subreq;
321 	fsreq->pos	= subreq->start + subreq->transferred;
322 	fsreq->len	= subreq->len   - subreq->transferred;
323 	fsreq->key	= key_get(subreq->rreq->netfs_priv);
324 	fsreq->vnode	= vnode;
325 	fsreq->iter	= &fsreq->def_iter;
326 
327 	iov_iter_xarray(&fsreq->def_iter, READ,
328 			&fsreq->vnode->netfs.inode.i_mapping->i_pages,
329 			fsreq->pos, fsreq->len);
330 
331 	afs_fetch_data(fsreq->vnode, fsreq);
332 	afs_put_read(fsreq);
333 }
334 
afs_symlink_read_folio(struct file * file,struct folio * folio)335 static int afs_symlink_read_folio(struct file *file, struct folio *folio)
336 {
337 	struct afs_vnode *vnode = AFS_FS_I(folio->mapping->host);
338 	struct afs_read *fsreq;
339 	int ret;
340 
341 	fsreq = afs_alloc_read(GFP_NOFS);
342 	if (!fsreq)
343 		return -ENOMEM;
344 
345 	fsreq->pos	= folio_pos(folio);
346 	fsreq->len	= folio_size(folio);
347 	fsreq->vnode	= vnode;
348 	fsreq->iter	= &fsreq->def_iter;
349 	iov_iter_xarray(&fsreq->def_iter, READ, &folio->mapping->i_pages,
350 			fsreq->pos, fsreq->len);
351 
352 	ret = afs_fetch_data(fsreq->vnode, fsreq);
353 	if (ret == 0)
354 		folio_mark_uptodate(folio);
355 	folio_unlock(folio);
356 	return ret;
357 }
358 
afs_init_request(struct netfs_io_request * rreq,struct file * file)359 static int afs_init_request(struct netfs_io_request *rreq, struct file *file)
360 {
361 	rreq->netfs_priv = key_get(afs_file_key(file));
362 	return 0;
363 }
364 
afs_begin_cache_operation(struct netfs_io_request * rreq)365 static int afs_begin_cache_operation(struct netfs_io_request *rreq)
366 {
367 #ifdef CONFIG_AFS_FSCACHE
368 	struct afs_vnode *vnode = AFS_FS_I(rreq->inode);
369 
370 	return fscache_begin_read_operation(&rreq->cache_resources,
371 					    afs_vnode_cache(vnode));
372 #else
373 	return -ENOBUFS;
374 #endif
375 }
376 
afs_check_write_begin(struct file * file,loff_t pos,unsigned len,struct folio ** foliop,void ** _fsdata)377 static int afs_check_write_begin(struct file *file, loff_t pos, unsigned len,
378 				 struct folio **foliop, void **_fsdata)
379 {
380 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
381 
382 	return test_bit(AFS_VNODE_DELETED, &vnode->flags) ? -ESTALE : 0;
383 }
384 
afs_free_request(struct netfs_io_request * rreq)385 static void afs_free_request(struct netfs_io_request *rreq)
386 {
387 	key_put(rreq->netfs_priv);
388 }
389 
390 const struct netfs_request_ops afs_req_ops = {
391 	.init_request		= afs_init_request,
392 	.free_request		= afs_free_request,
393 	.begin_cache_operation	= afs_begin_cache_operation,
394 	.check_write_begin	= afs_check_write_begin,
395 	.issue_read		= afs_issue_read,
396 };
397 
afs_write_inode(struct inode * inode,struct writeback_control * wbc)398 int afs_write_inode(struct inode *inode, struct writeback_control *wbc)
399 {
400 	fscache_unpin_writeback(wbc, afs_vnode_cache(AFS_FS_I(inode)));
401 	return 0;
402 }
403 
404 /*
405  * Adjust the dirty region of the page on truncation or full invalidation,
406  * getting rid of the markers altogether if the region is entirely invalidated.
407  */
afs_invalidate_dirty(struct folio * folio,size_t offset,size_t length)408 static void afs_invalidate_dirty(struct folio *folio, size_t offset,
409 				 size_t length)
410 {
411 	struct afs_vnode *vnode = AFS_FS_I(folio_inode(folio));
412 	unsigned long priv;
413 	unsigned int f, t, end = offset + length;
414 
415 	priv = (unsigned long)folio_get_private(folio);
416 
417 	/* we clean up only if the entire page is being invalidated */
418 	if (offset == 0 && length == folio_size(folio))
419 		goto full_invalidate;
420 
421 	 /* If the page was dirtied by page_mkwrite(), the PTE stays writable
422 	  * and we don't get another notification to tell us to expand it
423 	  * again.
424 	  */
425 	if (afs_is_folio_dirty_mmapped(priv))
426 		return;
427 
428 	/* We may need to shorten the dirty region */
429 	f = afs_folio_dirty_from(folio, priv);
430 	t = afs_folio_dirty_to(folio, priv);
431 
432 	if (t <= offset || f >= end)
433 		return; /* Doesn't overlap */
434 
435 	if (f < offset && t > end)
436 		return; /* Splits the dirty region - just absorb it */
437 
438 	if (f >= offset && t <= end)
439 		goto undirty;
440 
441 	if (f < offset)
442 		t = offset;
443 	else
444 		f = end;
445 	if (f == t)
446 		goto undirty;
447 
448 	priv = afs_folio_dirty(folio, f, t);
449 	folio_change_private(folio, (void *)priv);
450 	trace_afs_folio_dirty(vnode, tracepoint_string("trunc"), folio);
451 	return;
452 
453 undirty:
454 	trace_afs_folio_dirty(vnode, tracepoint_string("undirty"), folio);
455 	folio_clear_dirty_for_io(folio);
456 full_invalidate:
457 	trace_afs_folio_dirty(vnode, tracepoint_string("inval"), folio);
458 	folio_detach_private(folio);
459 }
460 
461 /*
462  * invalidate part or all of a page
463  * - release a page and clean up its private data if offset is 0 (indicating
464  *   the entire page)
465  */
afs_invalidate_folio(struct folio * folio,size_t offset,size_t length)466 static void afs_invalidate_folio(struct folio *folio, size_t offset,
467 			       size_t length)
468 {
469 	_enter("{%lu},%zu,%zu", folio->index, offset, length);
470 
471 	BUG_ON(!folio_test_locked(folio));
472 
473 	if (folio_get_private(folio))
474 		afs_invalidate_dirty(folio, offset, length);
475 
476 	folio_wait_fscache(folio);
477 	_leave("");
478 }
479 
480 /*
481  * release a page and clean up its private state if it's not busy
482  * - return true if the page can now be released, false if not
483  */
afs_release_folio(struct folio * folio,gfp_t gfp)484 static bool afs_release_folio(struct folio *folio, gfp_t gfp)
485 {
486 	struct afs_vnode *vnode = AFS_FS_I(folio_inode(folio));
487 
488 	_enter("{{%llx:%llu}[%lu],%lx},%x",
489 	       vnode->fid.vid, vnode->fid.vnode, folio_index(folio), folio->flags,
490 	       gfp);
491 
492 	/* deny if folio is being written to the cache and the caller hasn't
493 	 * elected to wait */
494 #ifdef CONFIG_AFS_FSCACHE
495 	if (folio_test_fscache(folio)) {
496 		if (current_is_kswapd() || !(gfp & __GFP_FS))
497 			return false;
498 		folio_wait_fscache(folio);
499 	}
500 	fscache_note_page_release(afs_vnode_cache(vnode));
501 #endif
502 
503 	if (folio_test_private(folio)) {
504 		trace_afs_folio_dirty(vnode, tracepoint_string("rel"), folio);
505 		folio_detach_private(folio);
506 	}
507 
508 	/* Indicate that the folio can be released */
509 	_leave(" = T");
510 	return true;
511 }
512 
afs_add_open_mmap(struct afs_vnode * vnode)513 static void afs_add_open_mmap(struct afs_vnode *vnode)
514 {
515 	if (atomic_inc_return(&vnode->cb_nr_mmap) == 1) {
516 		down_write(&vnode->volume->cell->fs_open_mmaps_lock);
517 
518 		if (list_empty(&vnode->cb_mmap_link))
519 			list_add_tail(&vnode->cb_mmap_link,
520 				      &vnode->volume->cell->fs_open_mmaps);
521 
522 		up_write(&vnode->volume->cell->fs_open_mmaps_lock);
523 	}
524 }
525 
afs_drop_open_mmap(struct afs_vnode * vnode)526 static void afs_drop_open_mmap(struct afs_vnode *vnode)
527 {
528 	if (!atomic_dec_and_test(&vnode->cb_nr_mmap))
529 		return;
530 
531 	down_write(&vnode->volume->cell->fs_open_mmaps_lock);
532 
533 	if (atomic_read(&vnode->cb_nr_mmap) == 0)
534 		list_del_init(&vnode->cb_mmap_link);
535 
536 	up_write(&vnode->volume->cell->fs_open_mmaps_lock);
537 	flush_work(&vnode->cb_work);
538 }
539 
540 /*
541  * Handle setting up a memory mapping on an AFS file.
542  */
afs_file_mmap(struct file * file,struct vm_area_struct * vma)543 static int afs_file_mmap(struct file *file, struct vm_area_struct *vma)
544 {
545 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
546 	int ret;
547 
548 	afs_add_open_mmap(vnode);
549 
550 	ret = generic_file_mmap(file, vma);
551 	if (ret == 0)
552 		vma->vm_ops = &afs_vm_ops;
553 	else
554 		afs_drop_open_mmap(vnode);
555 	return ret;
556 }
557 
afs_vm_open(struct vm_area_struct * vma)558 static void afs_vm_open(struct vm_area_struct *vma)
559 {
560 	afs_add_open_mmap(AFS_FS_I(file_inode(vma->vm_file)));
561 }
562 
afs_vm_close(struct vm_area_struct * vma)563 static void afs_vm_close(struct vm_area_struct *vma)
564 {
565 	afs_drop_open_mmap(AFS_FS_I(file_inode(vma->vm_file)));
566 }
567 
afs_vm_map_pages(struct vm_fault * vmf,pgoff_t start_pgoff,pgoff_t end_pgoff)568 static vm_fault_t afs_vm_map_pages(struct vm_fault *vmf, pgoff_t start_pgoff, pgoff_t end_pgoff)
569 {
570 	struct afs_vnode *vnode = AFS_FS_I(file_inode(vmf->vma->vm_file));
571 	struct afs_file *af = vmf->vma->vm_file->private_data;
572 
573 	switch (afs_validate(vnode, af->key)) {
574 	case 0:
575 		return filemap_map_pages(vmf, start_pgoff, end_pgoff);
576 	case -ENOMEM:
577 		return VM_FAULT_OOM;
578 	case -EINTR:
579 	case -ERESTARTSYS:
580 		return VM_FAULT_RETRY;
581 	case -ESTALE:
582 	default:
583 		return VM_FAULT_SIGBUS;
584 	}
585 }
586 
afs_file_read_iter(struct kiocb * iocb,struct iov_iter * iter)587 static ssize_t afs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
588 {
589 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
590 	struct afs_file *af = iocb->ki_filp->private_data;
591 	int ret;
592 
593 	ret = afs_validate(vnode, af->key);
594 	if (ret < 0)
595 		return ret;
596 
597 	return generic_file_read_iter(iocb, iter);
598 }
599