xref: /linux/fs/afs/dir.c (revision bdd1a21b52557ea8f61d0a5dc2f77151b576eb70)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* dir.c: AFS filesystem directory handling
3  *
4  * Copyright (C) 2002, 2018 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/fs.h>
10 #include <linux/namei.h>
11 #include <linux/pagemap.h>
12 #include <linux/swap.h>
13 #include <linux/ctype.h>
14 #include <linux/sched.h>
15 #include <linux/task_io_accounting_ops.h>
16 #include "internal.h"
17 #include "afs_fs.h"
18 #include "xdr_fs.h"
19 
20 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
21 				 unsigned int flags);
22 static int afs_dir_open(struct inode *inode, struct file *file);
23 static int afs_readdir(struct file *file, struct dir_context *ctx);
24 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags);
25 static int afs_d_delete(const struct dentry *dentry);
26 static void afs_d_iput(struct dentry *dentry, struct inode *inode);
27 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name, int nlen,
28 				  loff_t fpos, u64 ino, unsigned dtype);
29 static int afs_lookup_filldir(struct dir_context *ctx, const char *name, int nlen,
30 			      loff_t fpos, u64 ino, unsigned dtype);
31 static int afs_create(struct user_namespace *mnt_userns, struct inode *dir,
32 		      struct dentry *dentry, umode_t mode, bool excl);
33 static int afs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
34 		     struct dentry *dentry, umode_t mode);
35 static int afs_rmdir(struct inode *dir, struct dentry *dentry);
36 static int afs_unlink(struct inode *dir, struct dentry *dentry);
37 static int afs_link(struct dentry *from, struct inode *dir,
38 		    struct dentry *dentry);
39 static int afs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
40 		       struct dentry *dentry, const char *content);
41 static int afs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
42 		      struct dentry *old_dentry, struct inode *new_dir,
43 		      struct dentry *new_dentry, unsigned int flags);
44 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags);
45 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
46 				   unsigned int length);
47 
48 static int afs_dir_set_page_dirty(struct page *page)
49 {
50 	BUG(); /* This should never happen. */
51 }
52 
53 const struct file_operations afs_dir_file_operations = {
54 	.open		= afs_dir_open,
55 	.release	= afs_release,
56 	.iterate_shared	= afs_readdir,
57 	.lock		= afs_lock,
58 	.llseek		= generic_file_llseek,
59 };
60 
61 const struct inode_operations afs_dir_inode_operations = {
62 	.create		= afs_create,
63 	.lookup		= afs_lookup,
64 	.link		= afs_link,
65 	.unlink		= afs_unlink,
66 	.symlink	= afs_symlink,
67 	.mkdir		= afs_mkdir,
68 	.rmdir		= afs_rmdir,
69 	.rename		= afs_rename,
70 	.permission	= afs_permission,
71 	.getattr	= afs_getattr,
72 	.setattr	= afs_setattr,
73 };
74 
75 const struct address_space_operations afs_dir_aops = {
76 	.set_page_dirty	= afs_dir_set_page_dirty,
77 	.releasepage	= afs_dir_releasepage,
78 	.invalidatepage	= afs_dir_invalidatepage,
79 };
80 
81 const struct dentry_operations afs_fs_dentry_operations = {
82 	.d_revalidate	= afs_d_revalidate,
83 	.d_delete	= afs_d_delete,
84 	.d_release	= afs_d_release,
85 	.d_automount	= afs_d_automount,
86 	.d_iput		= afs_d_iput,
87 };
88 
89 struct afs_lookup_one_cookie {
90 	struct dir_context	ctx;
91 	struct qstr		name;
92 	bool			found;
93 	struct afs_fid		fid;
94 };
95 
96 struct afs_lookup_cookie {
97 	struct dir_context	ctx;
98 	struct qstr		name;
99 	bool			found;
100 	bool			one_only;
101 	unsigned short		nr_fids;
102 	struct afs_fid		fids[50];
103 };
104 
105 /*
106  * Drop the refs that we're holding on the pages we were reading into.  We've
107  * got refs on the first nr_pages pages.
108  */
109 static void afs_dir_read_cleanup(struct afs_read *req)
110 {
111 	struct address_space *mapping = req->vnode->vfs_inode.i_mapping;
112 	struct page *page;
113 	pgoff_t last = req->nr_pages - 1;
114 
115 	XA_STATE(xas, &mapping->i_pages, 0);
116 
117 	if (unlikely(!req->nr_pages))
118 		return;
119 
120 	rcu_read_lock();
121 	xas_for_each(&xas, page, last) {
122 		if (xas_retry(&xas, page))
123 			continue;
124 		BUG_ON(xa_is_value(page));
125 		BUG_ON(PageCompound(page));
126 		ASSERTCMP(page->mapping, ==, mapping);
127 
128 		put_page(page);
129 	}
130 
131 	rcu_read_unlock();
132 }
133 
134 /*
135  * check that a directory page is valid
136  */
137 static bool afs_dir_check_page(struct afs_vnode *dvnode, struct page *page,
138 			       loff_t i_size)
139 {
140 	struct afs_xdr_dir_page *dbuf;
141 	loff_t latter, off;
142 	int tmp, qty;
143 
144 	/* Determine how many magic numbers there should be in this page, but
145 	 * we must take care because the directory may change size under us.
146 	 */
147 	off = page_offset(page);
148 	if (i_size <= off)
149 		goto checked;
150 
151 	latter = i_size - off;
152 	if (latter >= PAGE_SIZE)
153 		qty = PAGE_SIZE;
154 	else
155 		qty = latter;
156 	qty /= sizeof(union afs_xdr_dir_block);
157 
158 	/* check them */
159 	dbuf = kmap_atomic(page);
160 	for (tmp = 0; tmp < qty; tmp++) {
161 		if (dbuf->blocks[tmp].hdr.magic != AFS_DIR_MAGIC) {
162 			printk("kAFS: %s(%lx): bad magic %d/%d is %04hx\n",
163 			       __func__, dvnode->vfs_inode.i_ino, tmp, qty,
164 			       ntohs(dbuf->blocks[tmp].hdr.magic));
165 			trace_afs_dir_check_failed(dvnode, off, i_size);
166 			kunmap(page);
167 			trace_afs_file_error(dvnode, -EIO, afs_file_error_dir_bad_magic);
168 			goto error;
169 		}
170 
171 		/* Make sure each block is NUL terminated so we can reasonably
172 		 * use string functions on it.  The filenames in the page
173 		 * *should* be NUL-terminated anyway.
174 		 */
175 		((u8 *)&dbuf->blocks[tmp])[AFS_DIR_BLOCK_SIZE - 1] = 0;
176 	}
177 
178 	kunmap_atomic(dbuf);
179 
180 checked:
181 	afs_stat_v(dvnode, n_read_dir);
182 	return true;
183 
184 error:
185 	return false;
186 }
187 
188 /*
189  * Dump the contents of a directory.
190  */
191 static void afs_dir_dump(struct afs_vnode *dvnode, struct afs_read *req)
192 {
193 	struct afs_xdr_dir_page *dbuf;
194 	struct address_space *mapping = dvnode->vfs_inode.i_mapping;
195 	struct page *page;
196 	unsigned int i, qty = PAGE_SIZE / sizeof(union afs_xdr_dir_block);
197 	pgoff_t last = req->nr_pages - 1;
198 
199 	XA_STATE(xas, &mapping->i_pages, 0);
200 
201 	pr_warn("DIR %llx:%llx f=%llx l=%llx al=%llx\n",
202 		dvnode->fid.vid, dvnode->fid.vnode,
203 		req->file_size, req->len, req->actual_len);
204 	pr_warn("DIR %llx %x %zx %zx\n",
205 		req->pos, req->nr_pages,
206 		req->iter->iov_offset,  iov_iter_count(req->iter));
207 
208 	xas_for_each(&xas, page, last) {
209 		if (xas_retry(&xas, page))
210 			continue;
211 
212 		BUG_ON(PageCompound(page));
213 		BUG_ON(page->mapping != mapping);
214 
215 		dbuf = kmap_atomic(page);
216 		for (i = 0; i < qty; i++) {
217 			union afs_xdr_dir_block *block = &dbuf->blocks[i];
218 
219 			pr_warn("[%02lx] %32phN\n", page->index * qty + i, block);
220 		}
221 		kunmap_atomic(dbuf);
222 	}
223 }
224 
225 /*
226  * Check all the pages in a directory.  All the pages are held pinned.
227  */
228 static int afs_dir_check(struct afs_vnode *dvnode, struct afs_read *req)
229 {
230 	struct address_space *mapping = dvnode->vfs_inode.i_mapping;
231 	struct page *page;
232 	pgoff_t last = req->nr_pages - 1;
233 	int ret = 0;
234 
235 	XA_STATE(xas, &mapping->i_pages, 0);
236 
237 	if (unlikely(!req->nr_pages))
238 		return 0;
239 
240 	rcu_read_lock();
241 	xas_for_each(&xas, page, last) {
242 		if (xas_retry(&xas, page))
243 			continue;
244 
245 		BUG_ON(PageCompound(page));
246 		BUG_ON(page->mapping != mapping);
247 
248 		if (!afs_dir_check_page(dvnode, page, req->file_size)) {
249 			afs_dir_dump(dvnode, req);
250 			ret = -EIO;
251 			break;
252 		}
253 	}
254 
255 	rcu_read_unlock();
256 	return ret;
257 }
258 
259 /*
260  * open an AFS directory file
261  */
262 static int afs_dir_open(struct inode *inode, struct file *file)
263 {
264 	_enter("{%lu}", inode->i_ino);
265 
266 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
267 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
268 
269 	if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(inode)->flags))
270 		return -ENOENT;
271 
272 	return afs_open(inode, file);
273 }
274 
275 /*
276  * Read the directory into the pagecache in one go, scrubbing the previous
277  * contents.  The list of pages is returned, pinning them so that they don't
278  * get reclaimed during the iteration.
279  */
280 static struct afs_read *afs_read_dir(struct afs_vnode *dvnode, struct key *key)
281 	__acquires(&dvnode->validate_lock)
282 {
283 	struct afs_read *req;
284 	loff_t i_size;
285 	int nr_pages, i, n;
286 	int ret;
287 
288 	_enter("");
289 
290 	req = kzalloc(sizeof(*req), GFP_KERNEL);
291 	if (!req)
292 		return ERR_PTR(-ENOMEM);
293 
294 	refcount_set(&req->usage, 1);
295 	req->vnode = dvnode;
296 	req->key = key_get(key);
297 	req->cleanup = afs_dir_read_cleanup;
298 
299 expand:
300 	i_size = i_size_read(&dvnode->vfs_inode);
301 	if (i_size < 2048) {
302 		ret = afs_bad(dvnode, afs_file_error_dir_small);
303 		goto error;
304 	}
305 	if (i_size > 2048 * 1024) {
306 		trace_afs_file_error(dvnode, -EFBIG, afs_file_error_dir_big);
307 		ret = -EFBIG;
308 		goto error;
309 	}
310 
311 	_enter("%llu", i_size);
312 
313 	nr_pages = (i_size + PAGE_SIZE - 1) / PAGE_SIZE;
314 
315 	req->actual_len = i_size; /* May change */
316 	req->len = nr_pages * PAGE_SIZE; /* We can ask for more than there is */
317 	req->data_version = dvnode->status.data_version; /* May change */
318 	iov_iter_xarray(&req->def_iter, READ, &dvnode->vfs_inode.i_mapping->i_pages,
319 			0, i_size);
320 	req->iter = &req->def_iter;
321 
322 	/* Fill in any gaps that we might find where the memory reclaimer has
323 	 * been at work and pin all the pages.  If there are any gaps, we will
324 	 * need to reread the entire directory contents.
325 	 */
326 	i = req->nr_pages;
327 	while (i < nr_pages) {
328 		struct page *pages[8], *page;
329 
330 		n = find_get_pages_contig(dvnode->vfs_inode.i_mapping, i,
331 					  min_t(unsigned int, nr_pages - i,
332 						ARRAY_SIZE(pages)),
333 					  pages);
334 		_debug("find %u at %u/%u", n, i, nr_pages);
335 
336 		if (n == 0) {
337 			gfp_t gfp = dvnode->vfs_inode.i_mapping->gfp_mask;
338 
339 			if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
340 				afs_stat_v(dvnode, n_inval);
341 
342 			ret = -ENOMEM;
343 			page = __page_cache_alloc(gfp);
344 			if (!page)
345 				goto error;
346 			ret = add_to_page_cache_lru(page,
347 						    dvnode->vfs_inode.i_mapping,
348 						    i, gfp);
349 			if (ret < 0)
350 				goto error;
351 
352 			attach_page_private(page, (void *)1);
353 			unlock_page(page);
354 			req->nr_pages++;
355 			i++;
356 		} else {
357 			req->nr_pages += n;
358 			i += n;
359 		}
360 	}
361 
362 	/* If we're going to reload, we need to lock all the pages to prevent
363 	 * races.
364 	 */
365 	ret = -ERESTARTSYS;
366 	if (down_read_killable(&dvnode->validate_lock) < 0)
367 		goto error;
368 
369 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
370 		goto success;
371 
372 	up_read(&dvnode->validate_lock);
373 	if (down_write_killable(&dvnode->validate_lock) < 0)
374 		goto error;
375 
376 	if (!test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
377 		trace_afs_reload_dir(dvnode);
378 		ret = afs_fetch_data(dvnode, req);
379 		if (ret < 0)
380 			goto error_unlock;
381 
382 		task_io_account_read(PAGE_SIZE * req->nr_pages);
383 
384 		if (req->len < req->file_size) {
385 			/* The content has grown, so we need to expand the
386 			 * buffer.
387 			 */
388 			up_write(&dvnode->validate_lock);
389 			goto expand;
390 		}
391 
392 		/* Validate the data we just read. */
393 		ret = afs_dir_check(dvnode, req);
394 		if (ret < 0)
395 			goto error_unlock;
396 
397 		// TODO: Trim excess pages
398 
399 		set_bit(AFS_VNODE_DIR_VALID, &dvnode->flags);
400 	}
401 
402 	downgrade_write(&dvnode->validate_lock);
403 success:
404 	return req;
405 
406 error_unlock:
407 	up_write(&dvnode->validate_lock);
408 error:
409 	afs_put_read(req);
410 	_leave(" = %d", ret);
411 	return ERR_PTR(ret);
412 }
413 
414 /*
415  * deal with one block in an AFS directory
416  */
417 static int afs_dir_iterate_block(struct afs_vnode *dvnode,
418 				 struct dir_context *ctx,
419 				 union afs_xdr_dir_block *block,
420 				 unsigned blkoff)
421 {
422 	union afs_xdr_dirent *dire;
423 	unsigned offset, next, curr, nr_slots;
424 	size_t nlen;
425 	int tmp;
426 
427 	_enter("%u,%x,%p,,",(unsigned)ctx->pos,blkoff,block);
428 
429 	curr = (ctx->pos - blkoff) / sizeof(union afs_xdr_dirent);
430 
431 	/* walk through the block, an entry at a time */
432 	for (offset = (blkoff == 0 ? AFS_DIR_RESV_BLOCKS0 : AFS_DIR_RESV_BLOCKS);
433 	     offset < AFS_DIR_SLOTS_PER_BLOCK;
434 	     offset = next
435 	     ) {
436 		/* skip entries marked unused in the bitmap */
437 		if (!(block->hdr.bitmap[offset / 8] &
438 		      (1 << (offset % 8)))) {
439 			_debug("ENT[%zu.%u]: unused",
440 			       blkoff / sizeof(union afs_xdr_dir_block), offset);
441 			next = offset + 1;
442 			if (offset >= curr)
443 				ctx->pos = blkoff +
444 					next * sizeof(union afs_xdr_dirent);
445 			continue;
446 		}
447 
448 		/* got a valid entry */
449 		dire = &block->dirents[offset];
450 		nlen = strnlen(dire->u.name,
451 			       sizeof(*block) -
452 			       offset * sizeof(union afs_xdr_dirent));
453 		if (nlen > AFSNAMEMAX - 1) {
454 			_debug("ENT[%zu]: name too long (len %u/%zu)",
455 			       blkoff / sizeof(union afs_xdr_dir_block),
456 			       offset, nlen);
457 			return afs_bad(dvnode, afs_file_error_dir_name_too_long);
458 		}
459 
460 		_debug("ENT[%zu.%u]: %s %zu \"%s\"",
461 		       blkoff / sizeof(union afs_xdr_dir_block), offset,
462 		       (offset < curr ? "skip" : "fill"),
463 		       nlen, dire->u.name);
464 
465 		nr_slots = afs_dir_calc_slots(nlen);
466 		next = offset + nr_slots;
467 		if (next > AFS_DIR_SLOTS_PER_BLOCK) {
468 			_debug("ENT[%zu.%u]:"
469 			       " %u extends beyond end dir block"
470 			       " (len %zu)",
471 			       blkoff / sizeof(union afs_xdr_dir_block),
472 			       offset, next, nlen);
473 			return afs_bad(dvnode, afs_file_error_dir_over_end);
474 		}
475 
476 		/* Check that the name-extension dirents are all allocated */
477 		for (tmp = 1; tmp < nr_slots; tmp++) {
478 			unsigned int ix = offset + tmp;
479 			if (!(block->hdr.bitmap[ix / 8] & (1 << (ix % 8)))) {
480 				_debug("ENT[%zu.u]:"
481 				       " %u unmarked extension (%u/%u)",
482 				       blkoff / sizeof(union afs_xdr_dir_block),
483 				       offset, tmp, nr_slots);
484 				return afs_bad(dvnode, afs_file_error_dir_unmarked_ext);
485 			}
486 		}
487 
488 		/* skip if starts before the current position */
489 		if (offset < curr)
490 			continue;
491 
492 		/* found the next entry */
493 		if (!dir_emit(ctx, dire->u.name, nlen,
494 			      ntohl(dire->u.vnode),
495 			      (ctx->actor == afs_lookup_filldir ||
496 			       ctx->actor == afs_lookup_one_filldir)?
497 			      ntohl(dire->u.unique) : DT_UNKNOWN)) {
498 			_leave(" = 0 [full]");
499 			return 0;
500 		}
501 
502 		ctx->pos = blkoff + next * sizeof(union afs_xdr_dirent);
503 	}
504 
505 	_leave(" = 1 [more]");
506 	return 1;
507 }
508 
509 /*
510  * iterate through the data blob that lists the contents of an AFS directory
511  */
512 static int afs_dir_iterate(struct inode *dir, struct dir_context *ctx,
513 			   struct key *key, afs_dataversion_t *_dir_version)
514 {
515 	struct afs_vnode *dvnode = AFS_FS_I(dir);
516 	struct afs_xdr_dir_page *dbuf;
517 	union afs_xdr_dir_block *dblock;
518 	struct afs_read *req;
519 	struct page *page;
520 	unsigned blkoff, limit;
521 	void __rcu **slot;
522 	int ret;
523 
524 	_enter("{%lu},%u,,", dir->i_ino, (unsigned)ctx->pos);
525 
526 	if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(dir)->flags)) {
527 		_leave(" = -ESTALE");
528 		return -ESTALE;
529 	}
530 
531 	req = afs_read_dir(dvnode, key);
532 	if (IS_ERR(req))
533 		return PTR_ERR(req);
534 	*_dir_version = req->data_version;
535 
536 	/* round the file position up to the next entry boundary */
537 	ctx->pos += sizeof(union afs_xdr_dirent) - 1;
538 	ctx->pos &= ~(sizeof(union afs_xdr_dirent) - 1);
539 
540 	/* walk through the blocks in sequence */
541 	ret = 0;
542 	while (ctx->pos < req->actual_len) {
543 		blkoff = ctx->pos & ~(sizeof(union afs_xdr_dir_block) - 1);
544 
545 		/* Fetch the appropriate page from the directory and re-add it
546 		 * to the LRU.  We have all the pages pinned with an extra ref.
547 		 */
548 		rcu_read_lock();
549 		page = NULL;
550 		slot = radix_tree_lookup_slot(&dvnode->vfs_inode.i_mapping->i_pages,
551 					      blkoff / PAGE_SIZE);
552 		if (slot)
553 			page = radix_tree_deref_slot(slot);
554 		rcu_read_unlock();
555 		if (!page) {
556 			ret = afs_bad(dvnode, afs_file_error_dir_missing_page);
557 			break;
558 		}
559 		mark_page_accessed(page);
560 
561 		limit = blkoff & ~(PAGE_SIZE - 1);
562 
563 		dbuf = kmap(page);
564 
565 		/* deal with the individual blocks stashed on this page */
566 		do {
567 			dblock = &dbuf->blocks[(blkoff % PAGE_SIZE) /
568 					       sizeof(union afs_xdr_dir_block)];
569 			ret = afs_dir_iterate_block(dvnode, ctx, dblock, blkoff);
570 			if (ret != 1) {
571 				kunmap(page);
572 				goto out;
573 			}
574 
575 			blkoff += sizeof(union afs_xdr_dir_block);
576 
577 		} while (ctx->pos < dir->i_size && blkoff < limit);
578 
579 		kunmap(page);
580 		ret = 0;
581 	}
582 
583 out:
584 	up_read(&dvnode->validate_lock);
585 	afs_put_read(req);
586 	_leave(" = %d", ret);
587 	return ret;
588 }
589 
590 /*
591  * read an AFS directory
592  */
593 static int afs_readdir(struct file *file, struct dir_context *ctx)
594 {
595 	afs_dataversion_t dir_version;
596 
597 	return afs_dir_iterate(file_inode(file), ctx, afs_file_key(file),
598 			       &dir_version);
599 }
600 
601 /*
602  * Search the directory for a single name
603  * - if afs_dir_iterate_block() spots this function, it'll pass the FID
604  *   uniquifier through dtype
605  */
606 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name,
607 				  int nlen, loff_t fpos, u64 ino, unsigned dtype)
608 {
609 	struct afs_lookup_one_cookie *cookie =
610 		container_of(ctx, struct afs_lookup_one_cookie, ctx);
611 
612 	_enter("{%s,%u},%s,%u,,%llu,%u",
613 	       cookie->name.name, cookie->name.len, name, nlen,
614 	       (unsigned long long) ino, dtype);
615 
616 	/* insanity checks first */
617 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
618 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
619 
620 	if (cookie->name.len != nlen ||
621 	    memcmp(cookie->name.name, name, nlen) != 0) {
622 		_leave(" = 0 [no]");
623 		return 0;
624 	}
625 
626 	cookie->fid.vnode = ino;
627 	cookie->fid.unique = dtype;
628 	cookie->found = 1;
629 
630 	_leave(" = -1 [found]");
631 	return -1;
632 }
633 
634 /*
635  * Do a lookup of a single name in a directory
636  * - just returns the FID the dentry name maps to if found
637  */
638 static int afs_do_lookup_one(struct inode *dir, struct dentry *dentry,
639 			     struct afs_fid *fid, struct key *key,
640 			     afs_dataversion_t *_dir_version)
641 {
642 	struct afs_super_info *as = dir->i_sb->s_fs_info;
643 	struct afs_lookup_one_cookie cookie = {
644 		.ctx.actor = afs_lookup_one_filldir,
645 		.name = dentry->d_name,
646 		.fid.vid = as->volume->vid
647 	};
648 	int ret;
649 
650 	_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
651 
652 	/* search the directory */
653 	ret = afs_dir_iterate(dir, &cookie.ctx, key, _dir_version);
654 	if (ret < 0) {
655 		_leave(" = %d [iter]", ret);
656 		return ret;
657 	}
658 
659 	if (!cookie.found) {
660 		_leave(" = -ENOENT [not found]");
661 		return -ENOENT;
662 	}
663 
664 	*fid = cookie.fid;
665 	_leave(" = 0 { vn=%llu u=%u }", fid->vnode, fid->unique);
666 	return 0;
667 }
668 
669 /*
670  * search the directory for a name
671  * - if afs_dir_iterate_block() spots this function, it'll pass the FID
672  *   uniquifier through dtype
673  */
674 static int afs_lookup_filldir(struct dir_context *ctx, const char *name,
675 			      int nlen, loff_t fpos, u64 ino, unsigned dtype)
676 {
677 	struct afs_lookup_cookie *cookie =
678 		container_of(ctx, struct afs_lookup_cookie, ctx);
679 	int ret;
680 
681 	_enter("{%s,%u},%s,%u,,%llu,%u",
682 	       cookie->name.name, cookie->name.len, name, nlen,
683 	       (unsigned long long) ino, dtype);
684 
685 	/* insanity checks first */
686 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
687 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
688 
689 	if (cookie->found) {
690 		if (cookie->nr_fids < 50) {
691 			cookie->fids[cookie->nr_fids].vnode	= ino;
692 			cookie->fids[cookie->nr_fids].unique	= dtype;
693 			cookie->nr_fids++;
694 		}
695 	} else if (cookie->name.len == nlen &&
696 		   memcmp(cookie->name.name, name, nlen) == 0) {
697 		cookie->fids[1].vnode	= ino;
698 		cookie->fids[1].unique	= dtype;
699 		cookie->found = 1;
700 		if (cookie->one_only)
701 			return -1;
702 	}
703 
704 	ret = cookie->nr_fids >= 50 ? -1 : 0;
705 	_leave(" = %d", ret);
706 	return ret;
707 }
708 
709 /*
710  * Deal with the result of a successful lookup operation.  Turn all the files
711  * into inodes and save the first one - which is the one we actually want.
712  */
713 static void afs_do_lookup_success(struct afs_operation *op)
714 {
715 	struct afs_vnode_param *vp;
716 	struct afs_vnode *vnode;
717 	struct inode *inode;
718 	u32 abort_code;
719 	int i;
720 
721 	_enter("");
722 
723 	for (i = 0; i < op->nr_files; i++) {
724 		switch (i) {
725 		case 0:
726 			vp = &op->file[0];
727 			abort_code = vp->scb.status.abort_code;
728 			if (abort_code != 0) {
729 				op->ac.abort_code = abort_code;
730 				op->error = afs_abort_to_error(abort_code);
731 			}
732 			break;
733 
734 		case 1:
735 			vp = &op->file[1];
736 			break;
737 
738 		default:
739 			vp = &op->more_files[i - 2];
740 			break;
741 		}
742 
743 		if (!vp->scb.have_status && !vp->scb.have_error)
744 			continue;
745 
746 		_debug("do [%u]", i);
747 		if (vp->vnode) {
748 			if (!test_bit(AFS_VNODE_UNSET, &vp->vnode->flags))
749 				afs_vnode_commit_status(op, vp);
750 		} else if (vp->scb.status.abort_code == 0) {
751 			inode = afs_iget(op, vp);
752 			if (!IS_ERR(inode)) {
753 				vnode = AFS_FS_I(inode);
754 				afs_cache_permit(vnode, op->key,
755 						 0 /* Assume vnode->cb_break is 0 */ +
756 						 op->cb_v_break,
757 						 &vp->scb);
758 				vp->vnode = vnode;
759 				vp->put_vnode = true;
760 			}
761 		} else {
762 			_debug("- abort %d %llx:%llx.%x",
763 			       vp->scb.status.abort_code,
764 			       vp->fid.vid, vp->fid.vnode, vp->fid.unique);
765 		}
766 	}
767 
768 	_leave("");
769 }
770 
771 static const struct afs_operation_ops afs_inline_bulk_status_operation = {
772 	.issue_afs_rpc	= afs_fs_inline_bulk_status,
773 	.issue_yfs_rpc	= yfs_fs_inline_bulk_status,
774 	.success	= afs_do_lookup_success,
775 };
776 
777 static const struct afs_operation_ops afs_lookup_fetch_status_operation = {
778 	.issue_afs_rpc	= afs_fs_fetch_status,
779 	.issue_yfs_rpc	= yfs_fs_fetch_status,
780 	.success	= afs_do_lookup_success,
781 	.aborted	= afs_check_for_remote_deletion,
782 };
783 
784 /*
785  * See if we know that the server we expect to use doesn't support
786  * FS.InlineBulkStatus.
787  */
788 static bool afs_server_supports_ibulk(struct afs_vnode *dvnode)
789 {
790 	struct afs_server_list *slist;
791 	struct afs_volume *volume = dvnode->volume;
792 	struct afs_server *server;
793 	bool ret = true;
794 	int i;
795 
796 	if (!test_bit(AFS_VOLUME_MAYBE_NO_IBULK, &volume->flags))
797 		return true;
798 
799 	rcu_read_lock();
800 	slist = rcu_dereference(volume->servers);
801 
802 	for (i = 0; i < slist->nr_servers; i++) {
803 		server = slist->servers[i].server;
804 		if (server == dvnode->cb_server) {
805 			if (test_bit(AFS_SERVER_FL_NO_IBULK, &server->flags))
806 				ret = false;
807 			break;
808 		}
809 	}
810 
811 	rcu_read_unlock();
812 	return ret;
813 }
814 
815 /*
816  * Do a lookup in a directory.  We make use of bulk lookup to query a slew of
817  * files in one go and create inodes for them.  The inode of the file we were
818  * asked for is returned.
819  */
820 static struct inode *afs_do_lookup(struct inode *dir, struct dentry *dentry,
821 				   struct key *key)
822 {
823 	struct afs_lookup_cookie *cookie;
824 	struct afs_vnode_param *vp;
825 	struct afs_operation *op;
826 	struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode;
827 	struct inode *inode = NULL, *ti;
828 	afs_dataversion_t data_version = READ_ONCE(dvnode->status.data_version);
829 	long ret;
830 	int i;
831 
832 	_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
833 
834 	cookie = kzalloc(sizeof(struct afs_lookup_cookie), GFP_KERNEL);
835 	if (!cookie)
836 		return ERR_PTR(-ENOMEM);
837 
838 	for (i = 0; i < ARRAY_SIZE(cookie->fids); i++)
839 		cookie->fids[i].vid = dvnode->fid.vid;
840 	cookie->ctx.actor = afs_lookup_filldir;
841 	cookie->name = dentry->d_name;
842 	cookie->nr_fids = 2; /* slot 0 is saved for the fid we actually want
843 			      * and slot 1 for the directory */
844 
845 	if (!afs_server_supports_ibulk(dvnode))
846 		cookie->one_only = true;
847 
848 	/* search the directory */
849 	ret = afs_dir_iterate(dir, &cookie->ctx, key, &data_version);
850 	if (ret < 0)
851 		goto out;
852 
853 	dentry->d_fsdata = (void *)(unsigned long)data_version;
854 
855 	ret = -ENOENT;
856 	if (!cookie->found)
857 		goto out;
858 
859 	/* Check to see if we already have an inode for the primary fid. */
860 	inode = ilookup5(dir->i_sb, cookie->fids[1].vnode,
861 			 afs_ilookup5_test_by_fid, &cookie->fids[1]);
862 	if (inode)
863 		goto out; /* We do */
864 
865 	/* Okay, we didn't find it.  We need to query the server - and whilst
866 	 * we're doing that, we're going to attempt to look up a bunch of other
867 	 * vnodes also.
868 	 */
869 	op = afs_alloc_operation(NULL, dvnode->volume);
870 	if (IS_ERR(op)) {
871 		ret = PTR_ERR(op);
872 		goto out;
873 	}
874 
875 	afs_op_set_vnode(op, 0, dvnode);
876 	afs_op_set_fid(op, 1, &cookie->fids[1]);
877 
878 	op->nr_files = cookie->nr_fids;
879 	_debug("nr_files %u", op->nr_files);
880 
881 	/* Need space for examining all the selected files */
882 	op->error = -ENOMEM;
883 	if (op->nr_files > 2) {
884 		op->more_files = kvcalloc(op->nr_files - 2,
885 					  sizeof(struct afs_vnode_param),
886 					  GFP_KERNEL);
887 		if (!op->more_files)
888 			goto out_op;
889 
890 		for (i = 2; i < op->nr_files; i++) {
891 			vp = &op->more_files[i - 2];
892 			vp->fid = cookie->fids[i];
893 
894 			/* Find any inodes that already exist and get their
895 			 * callback counters.
896 			 */
897 			ti = ilookup5_nowait(dir->i_sb, vp->fid.vnode,
898 					     afs_ilookup5_test_by_fid, &vp->fid);
899 			if (!IS_ERR_OR_NULL(ti)) {
900 				vnode = AFS_FS_I(ti);
901 				vp->dv_before = vnode->status.data_version;
902 				vp->cb_break_before = afs_calc_vnode_cb_break(vnode);
903 				vp->vnode = vnode;
904 				vp->put_vnode = true;
905 				vp->speculative = true; /* vnode not locked */
906 			}
907 		}
908 	}
909 
910 	/* Try FS.InlineBulkStatus first.  Abort codes for the individual
911 	 * lookups contained therein are stored in the reply without aborting
912 	 * the whole operation.
913 	 */
914 	op->error = -ENOTSUPP;
915 	if (!cookie->one_only) {
916 		op->ops = &afs_inline_bulk_status_operation;
917 		afs_begin_vnode_operation(op);
918 		afs_wait_for_operation(op);
919 	}
920 
921 	if (op->error == -ENOTSUPP) {
922 		/* We could try FS.BulkStatus next, but this aborts the entire
923 		 * op if any of the lookups fails - so, for the moment, revert
924 		 * to FS.FetchStatus for op->file[1].
925 		 */
926 		op->fetch_status.which = 1;
927 		op->ops = &afs_lookup_fetch_status_operation;
928 		afs_begin_vnode_operation(op);
929 		afs_wait_for_operation(op);
930 	}
931 	inode = ERR_PTR(op->error);
932 
933 out_op:
934 	if (op->error == 0) {
935 		inode = &op->file[1].vnode->vfs_inode;
936 		op->file[1].vnode = NULL;
937 	}
938 
939 	if (op->file[0].scb.have_status)
940 		dentry->d_fsdata = (void *)(unsigned long)op->file[0].scb.status.data_version;
941 	else
942 		dentry->d_fsdata = (void *)(unsigned long)op->file[0].dv_before;
943 	ret = afs_put_operation(op);
944 out:
945 	kfree(cookie);
946 	_leave("");
947 	return inode ?: ERR_PTR(ret);
948 }
949 
950 /*
951  * Look up an entry in a directory with @sys substitution.
952  */
953 static struct dentry *afs_lookup_atsys(struct inode *dir, struct dentry *dentry,
954 				       struct key *key)
955 {
956 	struct afs_sysnames *subs;
957 	struct afs_net *net = afs_i2net(dir);
958 	struct dentry *ret;
959 	char *buf, *p, *name;
960 	int len, i;
961 
962 	_enter("");
963 
964 	ret = ERR_PTR(-ENOMEM);
965 	p = buf = kmalloc(AFSNAMEMAX, GFP_KERNEL);
966 	if (!buf)
967 		goto out_p;
968 	if (dentry->d_name.len > 4) {
969 		memcpy(p, dentry->d_name.name, dentry->d_name.len - 4);
970 		p += dentry->d_name.len - 4;
971 	}
972 
973 	/* There is an ordered list of substitutes that we have to try. */
974 	read_lock(&net->sysnames_lock);
975 	subs = net->sysnames;
976 	refcount_inc(&subs->usage);
977 	read_unlock(&net->sysnames_lock);
978 
979 	for (i = 0; i < subs->nr; i++) {
980 		name = subs->subs[i];
981 		len = dentry->d_name.len - 4 + strlen(name);
982 		if (len >= AFSNAMEMAX) {
983 			ret = ERR_PTR(-ENAMETOOLONG);
984 			goto out_s;
985 		}
986 
987 		strcpy(p, name);
988 		ret = lookup_one_len(buf, dentry->d_parent, len);
989 		if (IS_ERR(ret) || d_is_positive(ret))
990 			goto out_s;
991 		dput(ret);
992 	}
993 
994 	/* We don't want to d_add() the @sys dentry here as we don't want to
995 	 * the cached dentry to hide changes to the sysnames list.
996 	 */
997 	ret = NULL;
998 out_s:
999 	afs_put_sysnames(subs);
1000 	kfree(buf);
1001 out_p:
1002 	key_put(key);
1003 	return ret;
1004 }
1005 
1006 /*
1007  * look up an entry in a directory
1008  */
1009 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
1010 				 unsigned int flags)
1011 {
1012 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1013 	struct afs_fid fid = {};
1014 	struct inode *inode;
1015 	struct dentry *d;
1016 	struct key *key;
1017 	int ret;
1018 
1019 	_enter("{%llx:%llu},%p{%pd},",
1020 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, dentry);
1021 
1022 	ASSERTCMP(d_inode(dentry), ==, NULL);
1023 
1024 	if (dentry->d_name.len >= AFSNAMEMAX) {
1025 		_leave(" = -ENAMETOOLONG");
1026 		return ERR_PTR(-ENAMETOOLONG);
1027 	}
1028 
1029 	if (test_bit(AFS_VNODE_DELETED, &dvnode->flags)) {
1030 		_leave(" = -ESTALE");
1031 		return ERR_PTR(-ESTALE);
1032 	}
1033 
1034 	key = afs_request_key(dvnode->volume->cell);
1035 	if (IS_ERR(key)) {
1036 		_leave(" = %ld [key]", PTR_ERR(key));
1037 		return ERR_CAST(key);
1038 	}
1039 
1040 	ret = afs_validate(dvnode, key);
1041 	if (ret < 0) {
1042 		key_put(key);
1043 		_leave(" = %d [val]", ret);
1044 		return ERR_PTR(ret);
1045 	}
1046 
1047 	if (dentry->d_name.len >= 4 &&
1048 	    dentry->d_name.name[dentry->d_name.len - 4] == '@' &&
1049 	    dentry->d_name.name[dentry->d_name.len - 3] == 's' &&
1050 	    dentry->d_name.name[dentry->d_name.len - 2] == 'y' &&
1051 	    dentry->d_name.name[dentry->d_name.len - 1] == 's')
1052 		return afs_lookup_atsys(dir, dentry, key);
1053 
1054 	afs_stat_v(dvnode, n_lookup);
1055 	inode = afs_do_lookup(dir, dentry, key);
1056 	key_put(key);
1057 	if (inode == ERR_PTR(-ENOENT))
1058 		inode = afs_try_auto_mntpt(dentry, dir);
1059 
1060 	if (!IS_ERR_OR_NULL(inode))
1061 		fid = AFS_FS_I(inode)->fid;
1062 
1063 	_debug("splice %p", dentry->d_inode);
1064 	d = d_splice_alias(inode, dentry);
1065 	if (!IS_ERR_OR_NULL(d)) {
1066 		d->d_fsdata = dentry->d_fsdata;
1067 		trace_afs_lookup(dvnode, &d->d_name, &fid);
1068 	} else {
1069 		trace_afs_lookup(dvnode, &dentry->d_name, &fid);
1070 	}
1071 	_leave("");
1072 	return d;
1073 }
1074 
1075 /*
1076  * Check the validity of a dentry under RCU conditions.
1077  */
1078 static int afs_d_revalidate_rcu(struct dentry *dentry)
1079 {
1080 	struct afs_vnode *dvnode, *vnode;
1081 	struct dentry *parent;
1082 	struct inode *dir, *inode;
1083 	long dir_version, de_version;
1084 
1085 	_enter("%p", dentry);
1086 
1087 	/* Check the parent directory is still valid first. */
1088 	parent = READ_ONCE(dentry->d_parent);
1089 	dir = d_inode_rcu(parent);
1090 	if (!dir)
1091 		return -ECHILD;
1092 	dvnode = AFS_FS_I(dir);
1093 	if (test_bit(AFS_VNODE_DELETED, &dvnode->flags))
1094 		return -ECHILD;
1095 
1096 	if (!afs_check_validity(dvnode))
1097 		return -ECHILD;
1098 
1099 	/* We only need to invalidate a dentry if the server's copy changed
1100 	 * behind our back.  If we made the change, it's no problem.  Note that
1101 	 * on a 32-bit system, we only have 32 bits in the dentry to store the
1102 	 * version.
1103 	 */
1104 	dir_version = (long)READ_ONCE(dvnode->status.data_version);
1105 	de_version = (long)READ_ONCE(dentry->d_fsdata);
1106 	if (de_version != dir_version) {
1107 		dir_version = (long)READ_ONCE(dvnode->invalid_before);
1108 		if (de_version - dir_version < 0)
1109 			return -ECHILD;
1110 	}
1111 
1112 	/* Check to see if the vnode referred to by the dentry still
1113 	 * has a callback.
1114 	 */
1115 	if (d_really_is_positive(dentry)) {
1116 		inode = d_inode_rcu(dentry);
1117 		if (inode) {
1118 			vnode = AFS_FS_I(inode);
1119 			if (!afs_check_validity(vnode))
1120 				return -ECHILD;
1121 		}
1122 	}
1123 
1124 	return 1; /* Still valid */
1125 }
1126 
1127 /*
1128  * check that a dentry lookup hit has found a valid entry
1129  * - NOTE! the hit can be a negative hit too, so we can't assume we have an
1130  *   inode
1131  */
1132 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags)
1133 {
1134 	struct afs_vnode *vnode, *dir;
1135 	struct afs_fid fid;
1136 	struct dentry *parent;
1137 	struct inode *inode;
1138 	struct key *key;
1139 	afs_dataversion_t dir_version, invalid_before;
1140 	long de_version;
1141 	int ret;
1142 
1143 	if (flags & LOOKUP_RCU)
1144 		return afs_d_revalidate_rcu(dentry);
1145 
1146 	if (d_really_is_positive(dentry)) {
1147 		vnode = AFS_FS_I(d_inode(dentry));
1148 		_enter("{v={%llx:%llu} n=%pd fl=%lx},",
1149 		       vnode->fid.vid, vnode->fid.vnode, dentry,
1150 		       vnode->flags);
1151 	} else {
1152 		_enter("{neg n=%pd}", dentry);
1153 	}
1154 
1155 	key = afs_request_key(AFS_FS_S(dentry->d_sb)->volume->cell);
1156 	if (IS_ERR(key))
1157 		key = NULL;
1158 
1159 	if (d_really_is_positive(dentry)) {
1160 		inode = d_inode(dentry);
1161 		if (inode) {
1162 			vnode = AFS_FS_I(inode);
1163 			afs_validate(vnode, key);
1164 			if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1165 				goto out_bad;
1166 		}
1167 	}
1168 
1169 	/* lock down the parent dentry so we can peer at it */
1170 	parent = dget_parent(dentry);
1171 	dir = AFS_FS_I(d_inode(parent));
1172 
1173 	/* validate the parent directory */
1174 	afs_validate(dir, key);
1175 
1176 	if (test_bit(AFS_VNODE_DELETED, &dir->flags)) {
1177 		_debug("%pd: parent dir deleted", dentry);
1178 		goto out_bad_parent;
1179 	}
1180 
1181 	/* We only need to invalidate a dentry if the server's copy changed
1182 	 * behind our back.  If we made the change, it's no problem.  Note that
1183 	 * on a 32-bit system, we only have 32 bits in the dentry to store the
1184 	 * version.
1185 	 */
1186 	dir_version = dir->status.data_version;
1187 	de_version = (long)dentry->d_fsdata;
1188 	if (de_version == (long)dir_version)
1189 		goto out_valid_noupdate;
1190 
1191 	invalid_before = dir->invalid_before;
1192 	if (de_version - (long)invalid_before >= 0)
1193 		goto out_valid;
1194 
1195 	_debug("dir modified");
1196 	afs_stat_v(dir, n_reval);
1197 
1198 	/* search the directory for this vnode */
1199 	ret = afs_do_lookup_one(&dir->vfs_inode, dentry, &fid, key, &dir_version);
1200 	switch (ret) {
1201 	case 0:
1202 		/* the filename maps to something */
1203 		if (d_really_is_negative(dentry))
1204 			goto out_bad_parent;
1205 		inode = d_inode(dentry);
1206 		if (is_bad_inode(inode)) {
1207 			printk("kAFS: afs_d_revalidate: %pd2 has bad inode\n",
1208 			       dentry);
1209 			goto out_bad_parent;
1210 		}
1211 
1212 		vnode = AFS_FS_I(inode);
1213 
1214 		/* if the vnode ID has changed, then the dirent points to a
1215 		 * different file */
1216 		if (fid.vnode != vnode->fid.vnode) {
1217 			_debug("%pd: dirent changed [%llu != %llu]",
1218 			       dentry, fid.vnode,
1219 			       vnode->fid.vnode);
1220 			goto not_found;
1221 		}
1222 
1223 		/* if the vnode ID uniqifier has changed, then the file has
1224 		 * been deleted and replaced, and the original vnode ID has
1225 		 * been reused */
1226 		if (fid.unique != vnode->fid.unique) {
1227 			_debug("%pd: file deleted (uq %u -> %u I:%u)",
1228 			       dentry, fid.unique,
1229 			       vnode->fid.unique,
1230 			       vnode->vfs_inode.i_generation);
1231 			write_seqlock(&vnode->cb_lock);
1232 			set_bit(AFS_VNODE_DELETED, &vnode->flags);
1233 			write_sequnlock(&vnode->cb_lock);
1234 			goto not_found;
1235 		}
1236 		goto out_valid;
1237 
1238 	case -ENOENT:
1239 		/* the filename is unknown */
1240 		_debug("%pd: dirent not found", dentry);
1241 		if (d_really_is_positive(dentry))
1242 			goto not_found;
1243 		goto out_valid;
1244 
1245 	default:
1246 		_debug("failed to iterate dir %pd: %d",
1247 		       parent, ret);
1248 		goto out_bad_parent;
1249 	}
1250 
1251 out_valid:
1252 	dentry->d_fsdata = (void *)(unsigned long)dir_version;
1253 out_valid_noupdate:
1254 	dput(parent);
1255 	key_put(key);
1256 	_leave(" = 1 [valid]");
1257 	return 1;
1258 
1259 	/* the dirent, if it exists, now points to a different vnode */
1260 not_found:
1261 	spin_lock(&dentry->d_lock);
1262 	dentry->d_flags |= DCACHE_NFSFS_RENAMED;
1263 	spin_unlock(&dentry->d_lock);
1264 
1265 out_bad_parent:
1266 	_debug("dropping dentry %pd2", dentry);
1267 	dput(parent);
1268 out_bad:
1269 	key_put(key);
1270 
1271 	_leave(" = 0 [bad]");
1272 	return 0;
1273 }
1274 
1275 /*
1276  * allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
1277  * sleep)
1278  * - called from dput() when d_count is going to 0.
1279  * - return 1 to request dentry be unhashed, 0 otherwise
1280  */
1281 static int afs_d_delete(const struct dentry *dentry)
1282 {
1283 	_enter("%pd", dentry);
1284 
1285 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1286 		goto zap;
1287 
1288 	if (d_really_is_positive(dentry) &&
1289 	    (test_bit(AFS_VNODE_DELETED,   &AFS_FS_I(d_inode(dentry))->flags) ||
1290 	     test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(d_inode(dentry))->flags)))
1291 		goto zap;
1292 
1293 	_leave(" = 0 [keep]");
1294 	return 0;
1295 
1296 zap:
1297 	_leave(" = 1 [zap]");
1298 	return 1;
1299 }
1300 
1301 /*
1302  * Clean up sillyrename files on dentry removal.
1303  */
1304 static void afs_d_iput(struct dentry *dentry, struct inode *inode)
1305 {
1306 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1307 		afs_silly_iput(dentry, inode);
1308 	iput(inode);
1309 }
1310 
1311 /*
1312  * handle dentry release
1313  */
1314 void afs_d_release(struct dentry *dentry)
1315 {
1316 	_enter("%pd", dentry);
1317 }
1318 
1319 void afs_check_for_remote_deletion(struct afs_operation *op)
1320 {
1321 	struct afs_vnode *vnode = op->file[0].vnode;
1322 
1323 	switch (op->ac.abort_code) {
1324 	case VNOVNODE:
1325 		set_bit(AFS_VNODE_DELETED, &vnode->flags);
1326 		afs_break_callback(vnode, afs_cb_break_for_deleted);
1327 	}
1328 }
1329 
1330 /*
1331  * Create a new inode for create/mkdir/symlink
1332  */
1333 static void afs_vnode_new_inode(struct afs_operation *op)
1334 {
1335 	struct afs_vnode_param *vp = &op->file[1];
1336 	struct afs_vnode *vnode;
1337 	struct inode *inode;
1338 
1339 	_enter("");
1340 
1341 	ASSERTCMP(op->error, ==, 0);
1342 
1343 	inode = afs_iget(op, vp);
1344 	if (IS_ERR(inode)) {
1345 		/* ENOMEM or EINTR at a really inconvenient time - just abandon
1346 		 * the new directory on the server.
1347 		 */
1348 		op->error = PTR_ERR(inode);
1349 		return;
1350 	}
1351 
1352 	vnode = AFS_FS_I(inode);
1353 	set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
1354 	if (!op->error)
1355 		afs_cache_permit(vnode, op->key, vnode->cb_break, &vp->scb);
1356 	d_instantiate(op->dentry, inode);
1357 }
1358 
1359 static void afs_create_success(struct afs_operation *op)
1360 {
1361 	_enter("op=%08x", op->debug_id);
1362 	op->ctime = op->file[0].scb.status.mtime_client;
1363 	afs_vnode_commit_status(op, &op->file[0]);
1364 	afs_update_dentry_version(op, &op->file[0], op->dentry);
1365 	afs_vnode_new_inode(op);
1366 }
1367 
1368 static void afs_create_edit_dir(struct afs_operation *op)
1369 {
1370 	struct afs_vnode_param *dvp = &op->file[0];
1371 	struct afs_vnode_param *vp = &op->file[1];
1372 	struct afs_vnode *dvnode = dvp->vnode;
1373 
1374 	_enter("op=%08x", op->debug_id);
1375 
1376 	down_write(&dvnode->validate_lock);
1377 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1378 	    dvnode->status.data_version == dvp->dv_before + dvp->dv_delta)
1379 		afs_edit_dir_add(dvnode, &op->dentry->d_name, &vp->fid,
1380 				 op->create.reason);
1381 	up_write(&dvnode->validate_lock);
1382 }
1383 
1384 static void afs_create_put(struct afs_operation *op)
1385 {
1386 	_enter("op=%08x", op->debug_id);
1387 
1388 	if (op->error)
1389 		d_drop(op->dentry);
1390 }
1391 
1392 static const struct afs_operation_ops afs_mkdir_operation = {
1393 	.issue_afs_rpc	= afs_fs_make_dir,
1394 	.issue_yfs_rpc	= yfs_fs_make_dir,
1395 	.success	= afs_create_success,
1396 	.aborted	= afs_check_for_remote_deletion,
1397 	.edit_dir	= afs_create_edit_dir,
1398 	.put		= afs_create_put,
1399 };
1400 
1401 /*
1402  * create a directory on an AFS filesystem
1403  */
1404 static int afs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1405 		     struct dentry *dentry, umode_t mode)
1406 {
1407 	struct afs_operation *op;
1408 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1409 
1410 	_enter("{%llx:%llu},{%pd},%ho",
1411 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1412 
1413 	op = afs_alloc_operation(NULL, dvnode->volume);
1414 	if (IS_ERR(op)) {
1415 		d_drop(dentry);
1416 		return PTR_ERR(op);
1417 	}
1418 
1419 	afs_op_set_vnode(op, 0, dvnode);
1420 	op->file[0].dv_delta = 1;
1421 	op->file[0].modification = true;
1422 	op->file[0].update_ctime = true;
1423 	op->dentry	= dentry;
1424 	op->create.mode	= S_IFDIR | mode;
1425 	op->create.reason = afs_edit_dir_for_mkdir;
1426 	op->ops		= &afs_mkdir_operation;
1427 	return afs_do_sync_operation(op);
1428 }
1429 
1430 /*
1431  * Remove a subdir from a directory.
1432  */
1433 static void afs_dir_remove_subdir(struct dentry *dentry)
1434 {
1435 	if (d_really_is_positive(dentry)) {
1436 		struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1437 
1438 		clear_nlink(&vnode->vfs_inode);
1439 		set_bit(AFS_VNODE_DELETED, &vnode->flags);
1440 		clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1441 		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
1442 	}
1443 }
1444 
1445 static void afs_rmdir_success(struct afs_operation *op)
1446 {
1447 	_enter("op=%08x", op->debug_id);
1448 	op->ctime = op->file[0].scb.status.mtime_client;
1449 	afs_vnode_commit_status(op, &op->file[0]);
1450 	afs_update_dentry_version(op, &op->file[0], op->dentry);
1451 }
1452 
1453 static void afs_rmdir_edit_dir(struct afs_operation *op)
1454 {
1455 	struct afs_vnode_param *dvp = &op->file[0];
1456 	struct afs_vnode *dvnode = dvp->vnode;
1457 
1458 	_enter("op=%08x", op->debug_id);
1459 	afs_dir_remove_subdir(op->dentry);
1460 
1461 	down_write(&dvnode->validate_lock);
1462 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1463 	    dvnode->status.data_version == dvp->dv_before + dvp->dv_delta)
1464 		afs_edit_dir_remove(dvnode, &op->dentry->d_name,
1465 				    afs_edit_dir_for_rmdir);
1466 	up_write(&dvnode->validate_lock);
1467 }
1468 
1469 static void afs_rmdir_put(struct afs_operation *op)
1470 {
1471 	_enter("op=%08x", op->debug_id);
1472 	if (op->file[1].vnode)
1473 		up_write(&op->file[1].vnode->rmdir_lock);
1474 }
1475 
1476 static const struct afs_operation_ops afs_rmdir_operation = {
1477 	.issue_afs_rpc	= afs_fs_remove_dir,
1478 	.issue_yfs_rpc	= yfs_fs_remove_dir,
1479 	.success	= afs_rmdir_success,
1480 	.aborted	= afs_check_for_remote_deletion,
1481 	.edit_dir	= afs_rmdir_edit_dir,
1482 	.put		= afs_rmdir_put,
1483 };
1484 
1485 /*
1486  * remove a directory from an AFS filesystem
1487  */
1488 static int afs_rmdir(struct inode *dir, struct dentry *dentry)
1489 {
1490 	struct afs_operation *op;
1491 	struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode = NULL;
1492 	int ret;
1493 
1494 	_enter("{%llx:%llu},{%pd}",
1495 	       dvnode->fid.vid, dvnode->fid.vnode, dentry);
1496 
1497 	op = afs_alloc_operation(NULL, dvnode->volume);
1498 	if (IS_ERR(op))
1499 		return PTR_ERR(op);
1500 
1501 	afs_op_set_vnode(op, 0, dvnode);
1502 	op->file[0].dv_delta = 1;
1503 	op->file[0].modification = true;
1504 	op->file[0].update_ctime = true;
1505 
1506 	op->dentry	= dentry;
1507 	op->ops		= &afs_rmdir_operation;
1508 
1509 	/* Try to make sure we have a callback promise on the victim. */
1510 	if (d_really_is_positive(dentry)) {
1511 		vnode = AFS_FS_I(d_inode(dentry));
1512 		ret = afs_validate(vnode, op->key);
1513 		if (ret < 0)
1514 			goto error;
1515 	}
1516 
1517 	if (vnode) {
1518 		ret = down_write_killable(&vnode->rmdir_lock);
1519 		if (ret < 0)
1520 			goto error;
1521 		op->file[1].vnode = vnode;
1522 	}
1523 
1524 	return afs_do_sync_operation(op);
1525 
1526 error:
1527 	return afs_put_operation(op);
1528 }
1529 
1530 /*
1531  * Remove a link to a file or symlink from a directory.
1532  *
1533  * If the file was not deleted due to excess hard links, the fileserver will
1534  * break the callback promise on the file - if it had one - before it returns
1535  * to us, and if it was deleted, it won't
1536  *
1537  * However, if we didn't have a callback promise outstanding, or it was
1538  * outstanding on a different server, then it won't break it either...
1539  */
1540 static void afs_dir_remove_link(struct afs_operation *op)
1541 {
1542 	struct afs_vnode *dvnode = op->file[0].vnode;
1543 	struct afs_vnode *vnode = op->file[1].vnode;
1544 	struct dentry *dentry = op->dentry;
1545 	int ret;
1546 
1547 	if (op->error != 0 ||
1548 	    (op->file[1].scb.have_status && op->file[1].scb.have_error))
1549 		return;
1550 	if (d_really_is_positive(dentry))
1551 		return;
1552 
1553 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
1554 		/* Already done */
1555 	} else if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
1556 		write_seqlock(&vnode->cb_lock);
1557 		drop_nlink(&vnode->vfs_inode);
1558 		if (vnode->vfs_inode.i_nlink == 0) {
1559 			set_bit(AFS_VNODE_DELETED, &vnode->flags);
1560 			__afs_break_callback(vnode, afs_cb_break_for_unlink);
1561 		}
1562 		write_sequnlock(&vnode->cb_lock);
1563 	} else {
1564 		afs_break_callback(vnode, afs_cb_break_for_unlink);
1565 
1566 		if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1567 			_debug("AFS_VNODE_DELETED");
1568 
1569 		ret = afs_validate(vnode, op->key);
1570 		if (ret != -ESTALE)
1571 			op->error = ret;
1572 	}
1573 
1574 	_debug("nlink %d [val %d]", vnode->vfs_inode.i_nlink, op->error);
1575 }
1576 
1577 static void afs_unlink_success(struct afs_operation *op)
1578 {
1579 	_enter("op=%08x", op->debug_id);
1580 	op->ctime = op->file[0].scb.status.mtime_client;
1581 	afs_check_dir_conflict(op, &op->file[0]);
1582 	afs_vnode_commit_status(op, &op->file[0]);
1583 	afs_vnode_commit_status(op, &op->file[1]);
1584 	afs_update_dentry_version(op, &op->file[0], op->dentry);
1585 	afs_dir_remove_link(op);
1586 }
1587 
1588 static void afs_unlink_edit_dir(struct afs_operation *op)
1589 {
1590 	struct afs_vnode_param *dvp = &op->file[0];
1591 	struct afs_vnode *dvnode = dvp->vnode;
1592 
1593 	_enter("op=%08x", op->debug_id);
1594 	down_write(&dvnode->validate_lock);
1595 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1596 	    dvnode->status.data_version == dvp->dv_before + dvp->dv_delta)
1597 		afs_edit_dir_remove(dvnode, &op->dentry->d_name,
1598 				    afs_edit_dir_for_unlink);
1599 	up_write(&dvnode->validate_lock);
1600 }
1601 
1602 static void afs_unlink_put(struct afs_operation *op)
1603 {
1604 	_enter("op=%08x", op->debug_id);
1605 	if (op->unlink.need_rehash && op->error < 0 && op->error != -ENOENT)
1606 		d_rehash(op->dentry);
1607 }
1608 
1609 static const struct afs_operation_ops afs_unlink_operation = {
1610 	.issue_afs_rpc	= afs_fs_remove_file,
1611 	.issue_yfs_rpc	= yfs_fs_remove_file,
1612 	.success	= afs_unlink_success,
1613 	.aborted	= afs_check_for_remote_deletion,
1614 	.edit_dir	= afs_unlink_edit_dir,
1615 	.put		= afs_unlink_put,
1616 };
1617 
1618 /*
1619  * Remove a file or symlink from an AFS filesystem.
1620  */
1621 static int afs_unlink(struct inode *dir, struct dentry *dentry)
1622 {
1623 	struct afs_operation *op;
1624 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1625 	struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1626 	int ret;
1627 
1628 	_enter("{%llx:%llu},{%pd}",
1629 	       dvnode->fid.vid, dvnode->fid.vnode, dentry);
1630 
1631 	if (dentry->d_name.len >= AFSNAMEMAX)
1632 		return -ENAMETOOLONG;
1633 
1634 	op = afs_alloc_operation(NULL, dvnode->volume);
1635 	if (IS_ERR(op))
1636 		return PTR_ERR(op);
1637 
1638 	afs_op_set_vnode(op, 0, dvnode);
1639 	op->file[0].dv_delta = 1;
1640 	op->file[0].modification = true;
1641 	op->file[0].update_ctime = true;
1642 
1643 	/* Try to make sure we have a callback promise on the victim. */
1644 	ret = afs_validate(vnode, op->key);
1645 	if (ret < 0) {
1646 		op->error = ret;
1647 		goto error;
1648 	}
1649 
1650 	spin_lock(&dentry->d_lock);
1651 	if (d_count(dentry) > 1) {
1652 		spin_unlock(&dentry->d_lock);
1653 		/* Start asynchronous writeout of the inode */
1654 		write_inode_now(d_inode(dentry), 0);
1655 		op->error = afs_sillyrename(dvnode, vnode, dentry, op->key);
1656 		goto error;
1657 	}
1658 	if (!d_unhashed(dentry)) {
1659 		/* Prevent a race with RCU lookup. */
1660 		__d_drop(dentry);
1661 		op->unlink.need_rehash = true;
1662 	}
1663 	spin_unlock(&dentry->d_lock);
1664 
1665 	op->file[1].vnode = vnode;
1666 	op->file[1].update_ctime = true;
1667 	op->file[1].op_unlinked = true;
1668 	op->dentry	= dentry;
1669 	op->ops		= &afs_unlink_operation;
1670 	afs_begin_vnode_operation(op);
1671 	afs_wait_for_operation(op);
1672 
1673 	/* If there was a conflict with a third party, check the status of the
1674 	 * unlinked vnode.
1675 	 */
1676 	if (op->error == 0 && (op->flags & AFS_OPERATION_DIR_CONFLICT)) {
1677 		op->file[1].update_ctime = false;
1678 		op->fetch_status.which = 1;
1679 		op->ops = &afs_fetch_status_operation;
1680 		afs_begin_vnode_operation(op);
1681 		afs_wait_for_operation(op);
1682 	}
1683 
1684 	return afs_put_operation(op);
1685 
1686 error:
1687 	return afs_put_operation(op);
1688 }
1689 
1690 static const struct afs_operation_ops afs_create_operation = {
1691 	.issue_afs_rpc	= afs_fs_create_file,
1692 	.issue_yfs_rpc	= yfs_fs_create_file,
1693 	.success	= afs_create_success,
1694 	.aborted	= afs_check_for_remote_deletion,
1695 	.edit_dir	= afs_create_edit_dir,
1696 	.put		= afs_create_put,
1697 };
1698 
1699 /*
1700  * create a regular file on an AFS filesystem
1701  */
1702 static int afs_create(struct user_namespace *mnt_userns, struct inode *dir,
1703 		      struct dentry *dentry, umode_t mode, bool excl)
1704 {
1705 	struct afs_operation *op;
1706 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1707 	int ret = -ENAMETOOLONG;
1708 
1709 	_enter("{%llx:%llu},{%pd},%ho",
1710 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1711 
1712 	if (dentry->d_name.len >= AFSNAMEMAX)
1713 		goto error;
1714 
1715 	op = afs_alloc_operation(NULL, dvnode->volume);
1716 	if (IS_ERR(op)) {
1717 		ret = PTR_ERR(op);
1718 		goto error;
1719 	}
1720 
1721 	afs_op_set_vnode(op, 0, dvnode);
1722 	op->file[0].dv_delta = 1;
1723 	op->file[0].modification = true;
1724 	op->file[0].update_ctime = true;
1725 
1726 	op->dentry	= dentry;
1727 	op->create.mode	= S_IFREG | mode;
1728 	op->create.reason = afs_edit_dir_for_create;
1729 	op->ops		= &afs_create_operation;
1730 	return afs_do_sync_operation(op);
1731 
1732 error:
1733 	d_drop(dentry);
1734 	_leave(" = %d", ret);
1735 	return ret;
1736 }
1737 
1738 static void afs_link_success(struct afs_operation *op)
1739 {
1740 	struct afs_vnode_param *dvp = &op->file[0];
1741 	struct afs_vnode_param *vp = &op->file[1];
1742 
1743 	_enter("op=%08x", op->debug_id);
1744 	op->ctime = dvp->scb.status.mtime_client;
1745 	afs_vnode_commit_status(op, dvp);
1746 	afs_vnode_commit_status(op, vp);
1747 	afs_update_dentry_version(op, dvp, op->dentry);
1748 	if (op->dentry_2->d_parent == op->dentry->d_parent)
1749 		afs_update_dentry_version(op, dvp, op->dentry_2);
1750 	ihold(&vp->vnode->vfs_inode);
1751 	d_instantiate(op->dentry, &vp->vnode->vfs_inode);
1752 }
1753 
1754 static void afs_link_put(struct afs_operation *op)
1755 {
1756 	_enter("op=%08x", op->debug_id);
1757 	if (op->error)
1758 		d_drop(op->dentry);
1759 }
1760 
1761 static const struct afs_operation_ops afs_link_operation = {
1762 	.issue_afs_rpc	= afs_fs_link,
1763 	.issue_yfs_rpc	= yfs_fs_link,
1764 	.success	= afs_link_success,
1765 	.aborted	= afs_check_for_remote_deletion,
1766 	.edit_dir	= afs_create_edit_dir,
1767 	.put		= afs_link_put,
1768 };
1769 
1770 /*
1771  * create a hard link between files in an AFS filesystem
1772  */
1773 static int afs_link(struct dentry *from, struct inode *dir,
1774 		    struct dentry *dentry)
1775 {
1776 	struct afs_operation *op;
1777 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1778 	struct afs_vnode *vnode = AFS_FS_I(d_inode(from));
1779 	int ret = -ENAMETOOLONG;
1780 
1781 	_enter("{%llx:%llu},{%llx:%llu},{%pd}",
1782 	       vnode->fid.vid, vnode->fid.vnode,
1783 	       dvnode->fid.vid, dvnode->fid.vnode,
1784 	       dentry);
1785 
1786 	if (dentry->d_name.len >= AFSNAMEMAX)
1787 		goto error;
1788 
1789 	op = afs_alloc_operation(NULL, dvnode->volume);
1790 	if (IS_ERR(op)) {
1791 		ret = PTR_ERR(op);
1792 		goto error;
1793 	}
1794 
1795 	afs_op_set_vnode(op, 0, dvnode);
1796 	afs_op_set_vnode(op, 1, vnode);
1797 	op->file[0].dv_delta = 1;
1798 	op->file[0].modification = true;
1799 	op->file[0].update_ctime = true;
1800 	op->file[1].update_ctime = true;
1801 
1802 	op->dentry		= dentry;
1803 	op->dentry_2		= from;
1804 	op->ops			= &afs_link_operation;
1805 	op->create.reason	= afs_edit_dir_for_link;
1806 	return afs_do_sync_operation(op);
1807 
1808 error:
1809 	d_drop(dentry);
1810 	_leave(" = %d", ret);
1811 	return ret;
1812 }
1813 
1814 static const struct afs_operation_ops afs_symlink_operation = {
1815 	.issue_afs_rpc	= afs_fs_symlink,
1816 	.issue_yfs_rpc	= yfs_fs_symlink,
1817 	.success	= afs_create_success,
1818 	.aborted	= afs_check_for_remote_deletion,
1819 	.edit_dir	= afs_create_edit_dir,
1820 	.put		= afs_create_put,
1821 };
1822 
1823 /*
1824  * create a symlink in an AFS filesystem
1825  */
1826 static int afs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1827 		       struct dentry *dentry, const char *content)
1828 {
1829 	struct afs_operation *op;
1830 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1831 	int ret;
1832 
1833 	_enter("{%llx:%llu},{%pd},%s",
1834 	       dvnode->fid.vid, dvnode->fid.vnode, dentry,
1835 	       content);
1836 
1837 	ret = -ENAMETOOLONG;
1838 	if (dentry->d_name.len >= AFSNAMEMAX)
1839 		goto error;
1840 
1841 	ret = -EINVAL;
1842 	if (strlen(content) >= AFSPATHMAX)
1843 		goto error;
1844 
1845 	op = afs_alloc_operation(NULL, dvnode->volume);
1846 	if (IS_ERR(op)) {
1847 		ret = PTR_ERR(op);
1848 		goto error;
1849 	}
1850 
1851 	afs_op_set_vnode(op, 0, dvnode);
1852 	op->file[0].dv_delta = 1;
1853 
1854 	op->dentry		= dentry;
1855 	op->ops			= &afs_symlink_operation;
1856 	op->create.reason	= afs_edit_dir_for_symlink;
1857 	op->create.symlink	= content;
1858 	return afs_do_sync_operation(op);
1859 
1860 error:
1861 	d_drop(dentry);
1862 	_leave(" = %d", ret);
1863 	return ret;
1864 }
1865 
1866 static void afs_rename_success(struct afs_operation *op)
1867 {
1868 	_enter("op=%08x", op->debug_id);
1869 
1870 	op->ctime = op->file[0].scb.status.mtime_client;
1871 	afs_check_dir_conflict(op, &op->file[1]);
1872 	afs_vnode_commit_status(op, &op->file[0]);
1873 	if (op->file[1].vnode != op->file[0].vnode) {
1874 		op->ctime = op->file[1].scb.status.mtime_client;
1875 		afs_vnode_commit_status(op, &op->file[1]);
1876 	}
1877 }
1878 
1879 static void afs_rename_edit_dir(struct afs_operation *op)
1880 {
1881 	struct afs_vnode_param *orig_dvp = &op->file[0];
1882 	struct afs_vnode_param *new_dvp = &op->file[1];
1883 	struct afs_vnode *orig_dvnode = orig_dvp->vnode;
1884 	struct afs_vnode *new_dvnode = new_dvp->vnode;
1885 	struct afs_vnode *vnode = AFS_FS_I(d_inode(op->dentry));
1886 	struct dentry *old_dentry = op->dentry;
1887 	struct dentry *new_dentry = op->dentry_2;
1888 	struct inode *new_inode;
1889 
1890 	_enter("op=%08x", op->debug_id);
1891 
1892 	if (op->rename.rehash) {
1893 		d_rehash(op->rename.rehash);
1894 		op->rename.rehash = NULL;
1895 	}
1896 
1897 	down_write(&orig_dvnode->validate_lock);
1898 	if (test_bit(AFS_VNODE_DIR_VALID, &orig_dvnode->flags) &&
1899 	    orig_dvnode->status.data_version == orig_dvp->dv_before + orig_dvp->dv_delta)
1900 		afs_edit_dir_remove(orig_dvnode, &old_dentry->d_name,
1901 				    afs_edit_dir_for_rename_0);
1902 
1903 	if (new_dvnode != orig_dvnode) {
1904 		up_write(&orig_dvnode->validate_lock);
1905 		down_write(&new_dvnode->validate_lock);
1906 	}
1907 
1908 	if (test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags) &&
1909 	    new_dvnode->status.data_version == new_dvp->dv_before + new_dvp->dv_delta) {
1910 		if (!op->rename.new_negative)
1911 			afs_edit_dir_remove(new_dvnode, &new_dentry->d_name,
1912 					    afs_edit_dir_for_rename_1);
1913 
1914 		afs_edit_dir_add(new_dvnode, &new_dentry->d_name,
1915 				 &vnode->fid, afs_edit_dir_for_rename_2);
1916 	}
1917 
1918 	new_inode = d_inode(new_dentry);
1919 	if (new_inode) {
1920 		spin_lock(&new_inode->i_lock);
1921 		if (S_ISDIR(new_inode->i_mode))
1922 			clear_nlink(new_inode);
1923 		else if (new_inode->i_nlink > 0)
1924 			drop_nlink(new_inode);
1925 		spin_unlock(&new_inode->i_lock);
1926 	}
1927 
1928 	/* Now we can update d_fsdata on the dentries to reflect their
1929 	 * new parent's data_version.
1930 	 *
1931 	 * Note that if we ever implement RENAME_EXCHANGE, we'll have
1932 	 * to update both dentries with opposing dir versions.
1933 	 */
1934 	afs_update_dentry_version(op, new_dvp, op->dentry);
1935 	afs_update_dentry_version(op, new_dvp, op->dentry_2);
1936 
1937 	d_move(old_dentry, new_dentry);
1938 
1939 	up_write(&new_dvnode->validate_lock);
1940 }
1941 
1942 static void afs_rename_put(struct afs_operation *op)
1943 {
1944 	_enter("op=%08x", op->debug_id);
1945 	if (op->rename.rehash)
1946 		d_rehash(op->rename.rehash);
1947 	dput(op->rename.tmp);
1948 	if (op->error)
1949 		d_rehash(op->dentry);
1950 }
1951 
1952 static const struct afs_operation_ops afs_rename_operation = {
1953 	.issue_afs_rpc	= afs_fs_rename,
1954 	.issue_yfs_rpc	= yfs_fs_rename,
1955 	.success	= afs_rename_success,
1956 	.edit_dir	= afs_rename_edit_dir,
1957 	.put		= afs_rename_put,
1958 };
1959 
1960 /*
1961  * rename a file in an AFS filesystem and/or move it between directories
1962  */
1963 static int afs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
1964 		      struct dentry *old_dentry, struct inode *new_dir,
1965 		      struct dentry *new_dentry, unsigned int flags)
1966 {
1967 	struct afs_operation *op;
1968 	struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
1969 	int ret;
1970 
1971 	if (flags)
1972 		return -EINVAL;
1973 
1974 	/* Don't allow silly-rename files be moved around. */
1975 	if (old_dentry->d_flags & DCACHE_NFSFS_RENAMED)
1976 		return -EINVAL;
1977 
1978 	vnode = AFS_FS_I(d_inode(old_dentry));
1979 	orig_dvnode = AFS_FS_I(old_dir);
1980 	new_dvnode = AFS_FS_I(new_dir);
1981 
1982 	_enter("{%llx:%llu},{%llx:%llu},{%llx:%llu},{%pd}",
1983 	       orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
1984 	       vnode->fid.vid, vnode->fid.vnode,
1985 	       new_dvnode->fid.vid, new_dvnode->fid.vnode,
1986 	       new_dentry);
1987 
1988 	op = afs_alloc_operation(NULL, orig_dvnode->volume);
1989 	if (IS_ERR(op))
1990 		return PTR_ERR(op);
1991 
1992 	afs_op_set_vnode(op, 0, orig_dvnode);
1993 	afs_op_set_vnode(op, 1, new_dvnode); /* May be same as orig_dvnode */
1994 	op->file[0].dv_delta = 1;
1995 	op->file[1].dv_delta = 1;
1996 	op->file[0].modification = true;
1997 	op->file[1].modification = true;
1998 	op->file[0].update_ctime = true;
1999 	op->file[1].update_ctime = true;
2000 
2001 	op->dentry		= old_dentry;
2002 	op->dentry_2		= new_dentry;
2003 	op->rename.new_negative	= d_is_negative(new_dentry);
2004 	op->ops			= &afs_rename_operation;
2005 
2006 	/* For non-directories, check whether the target is busy and if so,
2007 	 * make a copy of the dentry and then do a silly-rename.  If the
2008 	 * silly-rename succeeds, the copied dentry is hashed and becomes the
2009 	 * new target.
2010 	 */
2011 	if (d_is_positive(new_dentry) && !d_is_dir(new_dentry)) {
2012 		/* To prevent any new references to the target during the
2013 		 * rename, we unhash the dentry in advance.
2014 		 */
2015 		if (!d_unhashed(new_dentry)) {
2016 			d_drop(new_dentry);
2017 			op->rename.rehash = new_dentry;
2018 		}
2019 
2020 		if (d_count(new_dentry) > 2) {
2021 			/* copy the target dentry's name */
2022 			op->rename.tmp = d_alloc(new_dentry->d_parent,
2023 						 &new_dentry->d_name);
2024 			if (!op->rename.tmp) {
2025 				op->error = -ENOMEM;
2026 				goto error;
2027 			}
2028 
2029 			ret = afs_sillyrename(new_dvnode,
2030 					      AFS_FS_I(d_inode(new_dentry)),
2031 					      new_dentry, op->key);
2032 			if (ret) {
2033 				op->error = ret;
2034 				goto error;
2035 			}
2036 
2037 			op->dentry_2 = op->rename.tmp;
2038 			op->rename.rehash = NULL;
2039 			op->rename.new_negative = true;
2040 		}
2041 	}
2042 
2043 	/* This bit is potentially nasty as there's a potential race with
2044 	 * afs_d_revalidate{,_rcu}().  We have to change d_fsdata on the dentry
2045 	 * to reflect it's new parent's new data_version after the op, but
2046 	 * d_revalidate may see old_dentry between the op having taken place
2047 	 * and the version being updated.
2048 	 *
2049 	 * So drop the old_dentry for now to make other threads go through
2050 	 * lookup instead - which we hold a lock against.
2051 	 */
2052 	d_drop(old_dentry);
2053 
2054 	return afs_do_sync_operation(op);
2055 
2056 error:
2057 	return afs_put_operation(op);
2058 }
2059 
2060 /*
2061  * Release a directory page and clean up its private state if it's not busy
2062  * - return true if the page can now be released, false if not
2063  */
2064 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags)
2065 {
2066 	struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
2067 
2068 	_enter("{{%llx:%llu}[%lu]}", dvnode->fid.vid, dvnode->fid.vnode, page->index);
2069 
2070 	detach_page_private(page);
2071 
2072 	/* The directory will need reloading. */
2073 	if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2074 		afs_stat_v(dvnode, n_relpg);
2075 	return 1;
2076 }
2077 
2078 /*
2079  * invalidate part or all of a page
2080  * - release a page and clean up its private data if offset is 0 (indicating
2081  *   the entire page)
2082  */
2083 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
2084 				   unsigned int length)
2085 {
2086 	struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
2087 
2088 	_enter("{%lu},%u,%u", page->index, offset, length);
2089 
2090 	BUG_ON(!PageLocked(page));
2091 
2092 	/* The directory will need reloading. */
2093 	if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2094 		afs_stat_v(dvnode, n_inval);
2095 
2096 	/* we clean up only if the entire page is being invalidated */
2097 	if (offset == 0 && length == thp_size(page))
2098 		detach_page_private(page);
2099 }
2100