xref: /openbmc/linux/fs/afs/dir.c (revision 27f937cc)
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 inode *dir, struct dentry *dentry, umode_t mode,
32 		      bool excl);
33 static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
34 static int afs_rmdir(struct inode *dir, struct dentry *dentry);
35 static int afs_unlink(struct inode *dir, struct dentry *dentry);
36 static int afs_link(struct dentry *from, struct inode *dir,
37 		    struct dentry *dentry);
38 static int afs_symlink(struct inode *dir, struct dentry *dentry,
39 		       const char *content);
40 static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
41 		      struct inode *new_dir, struct dentry *new_dentry,
42 		      unsigned int flags);
43 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags);
44 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
45 				   unsigned int length);
46 
47 static int afs_dir_set_page_dirty(struct page *page)
48 {
49 	BUG(); /* This should never happen. */
50 }
51 
52 const struct file_operations afs_dir_file_operations = {
53 	.open		= afs_dir_open,
54 	.release	= afs_release,
55 	.iterate_shared	= afs_readdir,
56 	.lock		= afs_lock,
57 	.llseek		= generic_file_llseek,
58 };
59 
60 const struct inode_operations afs_dir_inode_operations = {
61 	.create		= afs_create,
62 	.lookup		= afs_lookup,
63 	.link		= afs_link,
64 	.unlink		= afs_unlink,
65 	.symlink	= afs_symlink,
66 	.mkdir		= afs_mkdir,
67 	.rmdir		= afs_rmdir,
68 	.rename		= afs_rename,
69 	.permission	= afs_permission,
70 	.getattr	= afs_getattr,
71 	.setattr	= afs_setattr,
72 	.listxattr	= afs_listxattr,
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 inode		**inodes;
103 	struct afs_status_cb	*statuses;
104 	struct afs_fid		fids[50];
105 };
106 
107 /*
108  * check that a directory page is valid
109  */
110 static bool afs_dir_check_page(struct afs_vnode *dvnode, struct page *page,
111 			       loff_t i_size)
112 {
113 	struct afs_xdr_dir_page *dbuf;
114 	loff_t latter, off;
115 	int tmp, qty;
116 
117 	/* Determine how many magic numbers there should be in this page, but
118 	 * we must take care because the directory may change size under us.
119 	 */
120 	off = page_offset(page);
121 	if (i_size <= off)
122 		goto checked;
123 
124 	latter = i_size - off;
125 	if (latter >= PAGE_SIZE)
126 		qty = PAGE_SIZE;
127 	else
128 		qty = latter;
129 	qty /= sizeof(union afs_xdr_dir_block);
130 
131 	/* check them */
132 	dbuf = kmap(page);
133 	for (tmp = 0; tmp < qty; tmp++) {
134 		if (dbuf->blocks[tmp].hdr.magic != AFS_DIR_MAGIC) {
135 			printk("kAFS: %s(%lx): bad magic %d/%d is %04hx\n",
136 			       __func__, dvnode->vfs_inode.i_ino, tmp, qty,
137 			       ntohs(dbuf->blocks[tmp].hdr.magic));
138 			trace_afs_dir_check_failed(dvnode, off, i_size);
139 			kunmap(page);
140 			trace_afs_file_error(dvnode, -EIO, afs_file_error_dir_bad_magic);
141 			goto error;
142 		}
143 
144 		/* Make sure each block is NUL terminated so we can reasonably
145 		 * use string functions on it.  The filenames in the page
146 		 * *should* be NUL-terminated anyway.
147 		 */
148 		((u8 *)&dbuf->blocks[tmp])[AFS_DIR_BLOCK_SIZE - 1] = 0;
149 	}
150 
151 	kunmap(page);
152 
153 checked:
154 	afs_stat_v(dvnode, n_read_dir);
155 	return true;
156 
157 error:
158 	return false;
159 }
160 
161 /*
162  * Check the contents of a directory that we've just read.
163  */
164 static bool afs_dir_check_pages(struct afs_vnode *dvnode, struct afs_read *req)
165 {
166 	struct afs_xdr_dir_page *dbuf;
167 	unsigned int i, j, qty = PAGE_SIZE / sizeof(union afs_xdr_dir_block);
168 
169 	for (i = 0; i < req->nr_pages; i++)
170 		if (!afs_dir_check_page(dvnode, req->pages[i], req->actual_len))
171 			goto bad;
172 	return true;
173 
174 bad:
175 	pr_warn("DIR %llx:%llx f=%llx l=%llx al=%llx r=%llx\n",
176 		dvnode->fid.vid, dvnode->fid.vnode,
177 		req->file_size, req->len, req->actual_len, req->remain);
178 	pr_warn("DIR %llx %x %x %x\n",
179 		req->pos, req->index, req->nr_pages, req->offset);
180 
181 	for (i = 0; i < req->nr_pages; i++) {
182 		dbuf = kmap(req->pages[i]);
183 		for (j = 0; j < qty; j++) {
184 			union afs_xdr_dir_block *block = &dbuf->blocks[j];
185 
186 			pr_warn("[%02x] %32phN\n", i * qty + j, block);
187 		}
188 		kunmap(req->pages[i]);
189 	}
190 	return false;
191 }
192 
193 /*
194  * open an AFS directory file
195  */
196 static int afs_dir_open(struct inode *inode, struct file *file)
197 {
198 	_enter("{%lu}", inode->i_ino);
199 
200 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
201 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
202 
203 	if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(inode)->flags))
204 		return -ENOENT;
205 
206 	return afs_open(inode, file);
207 }
208 
209 /*
210  * Read the directory into the pagecache in one go, scrubbing the previous
211  * contents.  The list of pages is returned, pinning them so that they don't
212  * get reclaimed during the iteration.
213  */
214 static struct afs_read *afs_read_dir(struct afs_vnode *dvnode, struct key *key)
215 	__acquires(&dvnode->validate_lock)
216 {
217 	struct afs_read *req;
218 	loff_t i_size;
219 	int nr_pages, nr_inline, i, n;
220 	int ret = -ENOMEM;
221 
222 retry:
223 	i_size = i_size_read(&dvnode->vfs_inode);
224 	if (i_size < 2048)
225 		return ERR_PTR(afs_bad(dvnode, afs_file_error_dir_small));
226 	if (i_size > 2048 * 1024) {
227 		trace_afs_file_error(dvnode, -EFBIG, afs_file_error_dir_big);
228 		return ERR_PTR(-EFBIG);
229 	}
230 
231 	_enter("%llu", i_size);
232 
233 	/* Get a request record to hold the page list.  We want to hold it
234 	 * inline if we can, but we don't want to make an order 1 allocation.
235 	 */
236 	nr_pages = (i_size + PAGE_SIZE - 1) / PAGE_SIZE;
237 	nr_inline = nr_pages;
238 	if (nr_inline > (PAGE_SIZE - sizeof(*req)) / sizeof(struct page *))
239 		nr_inline = 0;
240 
241 	req = kzalloc(struct_size(req, array, nr_inline), GFP_KERNEL);
242 	if (!req)
243 		return ERR_PTR(-ENOMEM);
244 
245 	refcount_set(&req->usage, 1);
246 	req->nr_pages = nr_pages;
247 	req->actual_len = i_size; /* May change */
248 	req->len = nr_pages * PAGE_SIZE; /* We can ask for more than there is */
249 	req->data_version = dvnode->status.data_version; /* May change */
250 	if (nr_inline > 0) {
251 		req->pages = req->array;
252 	} else {
253 		req->pages = kcalloc(nr_pages, sizeof(struct page *),
254 				     GFP_KERNEL);
255 		if (!req->pages)
256 			goto error;
257 	}
258 
259 	/* Get a list of all the pages that hold or will hold the directory
260 	 * content.  We need to fill in any gaps that we might find where the
261 	 * memory reclaimer has been at work.  If there are any gaps, we will
262 	 * need to reread the entire directory contents.
263 	 */
264 	i = 0;
265 	do {
266 		n = find_get_pages_contig(dvnode->vfs_inode.i_mapping, i,
267 					  req->nr_pages - i,
268 					  req->pages + i);
269 		_debug("find %u at %u/%u", n, i, req->nr_pages);
270 		if (n == 0) {
271 			gfp_t gfp = dvnode->vfs_inode.i_mapping->gfp_mask;
272 
273 			if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
274 				afs_stat_v(dvnode, n_inval);
275 
276 			ret = -ENOMEM;
277 			req->pages[i] = __page_cache_alloc(gfp);
278 			if (!req->pages[i])
279 				goto error;
280 			ret = add_to_page_cache_lru(req->pages[i],
281 						    dvnode->vfs_inode.i_mapping,
282 						    i, gfp);
283 			if (ret < 0)
284 				goto error;
285 
286 			set_page_private(req->pages[i], 1);
287 			SetPagePrivate(req->pages[i]);
288 			unlock_page(req->pages[i]);
289 			i++;
290 		} else {
291 			i += n;
292 		}
293 	} while (i < req->nr_pages);
294 
295 	/* If we're going to reload, we need to lock all the pages to prevent
296 	 * races.
297 	 */
298 	ret = -ERESTARTSYS;
299 	if (down_read_killable(&dvnode->validate_lock) < 0)
300 		goto error;
301 
302 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
303 		goto success;
304 
305 	up_read(&dvnode->validate_lock);
306 	if (down_write_killable(&dvnode->validate_lock) < 0)
307 		goto error;
308 
309 	if (!test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
310 		trace_afs_reload_dir(dvnode);
311 		ret = afs_fetch_data(dvnode, key, req);
312 		if (ret < 0)
313 			goto error_unlock;
314 
315 		task_io_account_read(PAGE_SIZE * req->nr_pages);
316 
317 		if (req->len < req->file_size)
318 			goto content_has_grown;
319 
320 		/* Validate the data we just read. */
321 		ret = -EIO;
322 		if (!afs_dir_check_pages(dvnode, req))
323 			goto error_unlock;
324 
325 		// TODO: Trim excess pages
326 
327 		set_bit(AFS_VNODE_DIR_VALID, &dvnode->flags);
328 	}
329 
330 	downgrade_write(&dvnode->validate_lock);
331 success:
332 	return req;
333 
334 error_unlock:
335 	up_write(&dvnode->validate_lock);
336 error:
337 	afs_put_read(req);
338 	_leave(" = %d", ret);
339 	return ERR_PTR(ret);
340 
341 content_has_grown:
342 	up_write(&dvnode->validate_lock);
343 	afs_put_read(req);
344 	goto retry;
345 }
346 
347 /*
348  * deal with one block in an AFS directory
349  */
350 static int afs_dir_iterate_block(struct afs_vnode *dvnode,
351 				 struct dir_context *ctx,
352 				 union afs_xdr_dir_block *block,
353 				 unsigned blkoff)
354 {
355 	union afs_xdr_dirent *dire;
356 	unsigned offset, next, curr;
357 	size_t nlen;
358 	int tmp;
359 
360 	_enter("%u,%x,%p,,",(unsigned)ctx->pos,blkoff,block);
361 
362 	curr = (ctx->pos - blkoff) / sizeof(union afs_xdr_dirent);
363 
364 	/* walk through the block, an entry at a time */
365 	for (offset = (blkoff == 0 ? AFS_DIR_RESV_BLOCKS0 : AFS_DIR_RESV_BLOCKS);
366 	     offset < AFS_DIR_SLOTS_PER_BLOCK;
367 	     offset = next
368 	     ) {
369 		next = offset + 1;
370 
371 		/* skip entries marked unused in the bitmap */
372 		if (!(block->hdr.bitmap[offset / 8] &
373 		      (1 << (offset % 8)))) {
374 			_debug("ENT[%zu.%u]: unused",
375 			       blkoff / sizeof(union afs_xdr_dir_block), offset);
376 			if (offset >= curr)
377 				ctx->pos = blkoff +
378 					next * sizeof(union afs_xdr_dirent);
379 			continue;
380 		}
381 
382 		/* got a valid entry */
383 		dire = &block->dirents[offset];
384 		nlen = strnlen(dire->u.name,
385 			       sizeof(*block) -
386 			       offset * sizeof(union afs_xdr_dirent));
387 
388 		_debug("ENT[%zu.%u]: %s %zu \"%s\"",
389 		       blkoff / sizeof(union afs_xdr_dir_block), offset,
390 		       (offset < curr ? "skip" : "fill"),
391 		       nlen, dire->u.name);
392 
393 		/* work out where the next possible entry is */
394 		for (tmp = nlen; tmp > 15; tmp -= sizeof(union afs_xdr_dirent)) {
395 			if (next >= AFS_DIR_SLOTS_PER_BLOCK) {
396 				_debug("ENT[%zu.%u]:"
397 				       " %u travelled beyond end dir block"
398 				       " (len %u/%zu)",
399 				       blkoff / sizeof(union afs_xdr_dir_block),
400 				       offset, next, tmp, nlen);
401 				return afs_bad(dvnode, afs_file_error_dir_over_end);
402 			}
403 			if (!(block->hdr.bitmap[next / 8] &
404 			      (1 << (next % 8)))) {
405 				_debug("ENT[%zu.%u]:"
406 				       " %u unmarked extension (len %u/%zu)",
407 				       blkoff / sizeof(union afs_xdr_dir_block),
408 				       offset, next, tmp, nlen);
409 				return afs_bad(dvnode, afs_file_error_dir_unmarked_ext);
410 			}
411 
412 			_debug("ENT[%zu.%u]: ext %u/%zu",
413 			       blkoff / sizeof(union afs_xdr_dir_block),
414 			       next, tmp, nlen);
415 			next++;
416 		}
417 
418 		/* skip if starts before the current position */
419 		if (offset < curr)
420 			continue;
421 
422 		/* found the next entry */
423 		if (!dir_emit(ctx, dire->u.name, nlen,
424 			      ntohl(dire->u.vnode),
425 			      (ctx->actor == afs_lookup_filldir ||
426 			       ctx->actor == afs_lookup_one_filldir)?
427 			      ntohl(dire->u.unique) : DT_UNKNOWN)) {
428 			_leave(" = 0 [full]");
429 			return 0;
430 		}
431 
432 		ctx->pos = blkoff + next * sizeof(union afs_xdr_dirent);
433 	}
434 
435 	_leave(" = 1 [more]");
436 	return 1;
437 }
438 
439 /*
440  * iterate through the data blob that lists the contents of an AFS directory
441  */
442 static int afs_dir_iterate(struct inode *dir, struct dir_context *ctx,
443 			   struct key *key, afs_dataversion_t *_dir_version)
444 {
445 	struct afs_vnode *dvnode = AFS_FS_I(dir);
446 	struct afs_xdr_dir_page *dbuf;
447 	union afs_xdr_dir_block *dblock;
448 	struct afs_read *req;
449 	struct page *page;
450 	unsigned blkoff, limit;
451 	int ret;
452 
453 	_enter("{%lu},%u,,", dir->i_ino, (unsigned)ctx->pos);
454 
455 	if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(dir)->flags)) {
456 		_leave(" = -ESTALE");
457 		return -ESTALE;
458 	}
459 
460 	req = afs_read_dir(dvnode, key);
461 	if (IS_ERR(req))
462 		return PTR_ERR(req);
463 	*_dir_version = req->data_version;
464 
465 	/* round the file position up to the next entry boundary */
466 	ctx->pos += sizeof(union afs_xdr_dirent) - 1;
467 	ctx->pos &= ~(sizeof(union afs_xdr_dirent) - 1);
468 
469 	/* walk through the blocks in sequence */
470 	ret = 0;
471 	while (ctx->pos < req->actual_len) {
472 		blkoff = ctx->pos & ~(sizeof(union afs_xdr_dir_block) - 1);
473 
474 		/* Fetch the appropriate page from the directory and re-add it
475 		 * to the LRU.
476 		 */
477 		page = req->pages[blkoff / PAGE_SIZE];
478 		if (!page) {
479 			ret = afs_bad(dvnode, afs_file_error_dir_missing_page);
480 			break;
481 		}
482 		mark_page_accessed(page);
483 
484 		limit = blkoff & ~(PAGE_SIZE - 1);
485 
486 		dbuf = kmap(page);
487 
488 		/* deal with the individual blocks stashed on this page */
489 		do {
490 			dblock = &dbuf->blocks[(blkoff % PAGE_SIZE) /
491 					       sizeof(union afs_xdr_dir_block)];
492 			ret = afs_dir_iterate_block(dvnode, ctx, dblock, blkoff);
493 			if (ret != 1) {
494 				kunmap(page);
495 				goto out;
496 			}
497 
498 			blkoff += sizeof(union afs_xdr_dir_block);
499 
500 		} while (ctx->pos < dir->i_size && blkoff < limit);
501 
502 		kunmap(page);
503 		ret = 0;
504 	}
505 
506 out:
507 	up_read(&dvnode->validate_lock);
508 	afs_put_read(req);
509 	_leave(" = %d", ret);
510 	return ret;
511 }
512 
513 /*
514  * read an AFS directory
515  */
516 static int afs_readdir(struct file *file, struct dir_context *ctx)
517 {
518 	afs_dataversion_t dir_version;
519 
520 	return afs_dir_iterate(file_inode(file), ctx, afs_file_key(file),
521 			       &dir_version);
522 }
523 
524 /*
525  * Search the directory for a single name
526  * - if afs_dir_iterate_block() spots this function, it'll pass the FID
527  *   uniquifier through dtype
528  */
529 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name,
530 				  int nlen, loff_t fpos, u64 ino, unsigned dtype)
531 {
532 	struct afs_lookup_one_cookie *cookie =
533 		container_of(ctx, struct afs_lookup_one_cookie, ctx);
534 
535 	_enter("{%s,%u},%s,%u,,%llu,%u",
536 	       cookie->name.name, cookie->name.len, name, nlen,
537 	       (unsigned long long) ino, dtype);
538 
539 	/* insanity checks first */
540 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
541 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
542 
543 	if (cookie->name.len != nlen ||
544 	    memcmp(cookie->name.name, name, nlen) != 0) {
545 		_leave(" = 0 [no]");
546 		return 0;
547 	}
548 
549 	cookie->fid.vnode = ino;
550 	cookie->fid.unique = dtype;
551 	cookie->found = 1;
552 
553 	_leave(" = -1 [found]");
554 	return -1;
555 }
556 
557 /*
558  * Do a lookup of a single name in a directory
559  * - just returns the FID the dentry name maps to if found
560  */
561 static int afs_do_lookup_one(struct inode *dir, struct dentry *dentry,
562 			     struct afs_fid *fid, struct key *key,
563 			     afs_dataversion_t *_dir_version)
564 {
565 	struct afs_super_info *as = dir->i_sb->s_fs_info;
566 	struct afs_lookup_one_cookie cookie = {
567 		.ctx.actor = afs_lookup_one_filldir,
568 		.name = dentry->d_name,
569 		.fid.vid = as->volume->vid
570 	};
571 	int ret;
572 
573 	_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
574 
575 	/* search the directory */
576 	ret = afs_dir_iterate(dir, &cookie.ctx, key, _dir_version);
577 	if (ret < 0) {
578 		_leave(" = %d [iter]", ret);
579 		return ret;
580 	}
581 
582 	ret = -ENOENT;
583 	if (!cookie.found) {
584 		_leave(" = -ENOENT [not found]");
585 		return -ENOENT;
586 	}
587 
588 	*fid = cookie.fid;
589 	_leave(" = 0 { vn=%llu u=%u }", fid->vnode, fid->unique);
590 	return 0;
591 }
592 
593 /*
594  * search the directory for a name
595  * - if afs_dir_iterate_block() spots this function, it'll pass the FID
596  *   uniquifier through dtype
597  */
598 static int afs_lookup_filldir(struct dir_context *ctx, const char *name,
599 			      int nlen, loff_t fpos, u64 ino, unsigned dtype)
600 {
601 	struct afs_lookup_cookie *cookie =
602 		container_of(ctx, struct afs_lookup_cookie, ctx);
603 	int ret;
604 
605 	_enter("{%s,%u},%s,%u,,%llu,%u",
606 	       cookie->name.name, cookie->name.len, name, nlen,
607 	       (unsigned long long) ino, dtype);
608 
609 	/* insanity checks first */
610 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
611 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
612 
613 	if (cookie->found) {
614 		if (cookie->nr_fids < 50) {
615 			cookie->fids[cookie->nr_fids].vnode	= ino;
616 			cookie->fids[cookie->nr_fids].unique	= dtype;
617 			cookie->nr_fids++;
618 		}
619 	} else if (cookie->name.len == nlen &&
620 		   memcmp(cookie->name.name, name, nlen) == 0) {
621 		cookie->fids[0].vnode	= ino;
622 		cookie->fids[0].unique	= dtype;
623 		cookie->found = 1;
624 		if (cookie->one_only)
625 			return -1;
626 	}
627 
628 	ret = cookie->nr_fids >= 50 ? -1 : 0;
629 	_leave(" = %d", ret);
630 	return ret;
631 }
632 
633 /*
634  * Do a lookup in a directory.  We make use of bulk lookup to query a slew of
635  * files in one go and create inodes for them.  The inode of the file we were
636  * asked for is returned.
637  */
638 static struct inode *afs_do_lookup(struct inode *dir, struct dentry *dentry,
639 				   struct key *key)
640 {
641 	struct afs_lookup_cookie *cookie;
642 	struct afs_cb_interest *dcbi, *cbi = NULL;
643 	struct afs_super_info *as = dir->i_sb->s_fs_info;
644 	struct afs_status_cb *scb;
645 	struct afs_iget_data iget_data;
646 	struct afs_fs_cursor fc;
647 	struct afs_server *server;
648 	struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode;
649 	struct inode *inode = NULL, *ti;
650 	afs_dataversion_t data_version = READ_ONCE(dvnode->status.data_version);
651 	int ret, i;
652 
653 	_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
654 
655 	cookie = kzalloc(sizeof(struct afs_lookup_cookie), GFP_KERNEL);
656 	if (!cookie)
657 		return ERR_PTR(-ENOMEM);
658 
659 	cookie->ctx.actor = afs_lookup_filldir;
660 	cookie->name = dentry->d_name;
661 	cookie->nr_fids = 1; /* slot 0 is saved for the fid we actually want */
662 
663 	read_seqlock_excl(&dvnode->cb_lock);
664 	dcbi = rcu_dereference_protected(dvnode->cb_interest,
665 					 lockdep_is_held(&dvnode->cb_lock.lock));
666 	if (dcbi) {
667 		server = dcbi->server;
668 		if (server &&
669 		    test_bit(AFS_SERVER_FL_NO_IBULK, &server->flags))
670 			cookie->one_only = true;
671 	}
672 	read_sequnlock_excl(&dvnode->cb_lock);
673 
674 	for (i = 0; i < 50; i++)
675 		cookie->fids[i].vid = as->volume->vid;
676 
677 	/* search the directory */
678 	ret = afs_dir_iterate(dir, &cookie->ctx, key, &data_version);
679 	if (ret < 0) {
680 		inode = ERR_PTR(ret);
681 		goto out;
682 	}
683 
684 	dentry->d_fsdata = (void *)(unsigned long)data_version;
685 
686 	inode = ERR_PTR(-ENOENT);
687 	if (!cookie->found)
688 		goto out;
689 
690 	/* Check to see if we already have an inode for the primary fid. */
691 	iget_data.fid = cookie->fids[0];
692 	iget_data.volume = dvnode->volume;
693 	iget_data.cb_v_break = dvnode->volume->cb_v_break;
694 	iget_data.cb_s_break = 0;
695 	inode = ilookup5(dir->i_sb, cookie->fids[0].vnode,
696 			 afs_iget5_test, &iget_data);
697 	if (inode)
698 		goto out;
699 
700 	/* Need space for examining all the selected files */
701 	inode = ERR_PTR(-ENOMEM);
702 	cookie->statuses = kvcalloc(cookie->nr_fids, sizeof(struct afs_status_cb),
703 				    GFP_KERNEL);
704 	if (!cookie->statuses)
705 		goto out;
706 
707 	cookie->inodes = kcalloc(cookie->nr_fids, sizeof(struct inode *),
708 				 GFP_KERNEL);
709 	if (!cookie->inodes)
710 		goto out_s;
711 
712 	for (i = 1; i < cookie->nr_fids; i++) {
713 		scb = &cookie->statuses[i];
714 
715 		/* Find any inodes that already exist and get their
716 		 * callback counters.
717 		 */
718 		iget_data.fid = cookie->fids[i];
719 		ti = ilookup5_nowait(dir->i_sb, iget_data.fid.vnode,
720 				     afs_iget5_test, &iget_data);
721 		if (!IS_ERR_OR_NULL(ti)) {
722 			vnode = AFS_FS_I(ti);
723 			scb->cb_break = afs_calc_vnode_cb_break(vnode);
724 			cookie->inodes[i] = ti;
725 		}
726 	}
727 
728 	/* Try FS.InlineBulkStatus first.  Abort codes for the individual
729 	 * lookups contained therein are stored in the reply without aborting
730 	 * the whole operation.
731 	 */
732 	if (cookie->one_only)
733 		goto no_inline_bulk_status;
734 
735 	inode = ERR_PTR(-ERESTARTSYS);
736 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
737 		while (afs_select_fileserver(&fc)) {
738 			if (test_bit(AFS_SERVER_FL_NO_IBULK,
739 				      &fc.cbi->server->flags)) {
740 				fc.ac.abort_code = RX_INVALID_OPERATION;
741 				fc.ac.error = -ECONNABORTED;
742 				break;
743 			}
744 			iget_data.cb_v_break = dvnode->volume->cb_v_break;
745 			iget_data.cb_s_break = fc.cbi->server->cb_s_break;
746 			afs_fs_inline_bulk_status(&fc,
747 						  afs_v2net(dvnode),
748 						  cookie->fids,
749 						  cookie->statuses,
750 						  cookie->nr_fids, NULL);
751 		}
752 
753 		if (fc.ac.error == 0)
754 			cbi = afs_get_cb_interest(fc.cbi);
755 		if (fc.ac.abort_code == RX_INVALID_OPERATION)
756 			set_bit(AFS_SERVER_FL_NO_IBULK, &fc.cbi->server->flags);
757 		inode = ERR_PTR(afs_end_vnode_operation(&fc));
758 	}
759 
760 	if (!IS_ERR(inode))
761 		goto success;
762 	if (fc.ac.abort_code != RX_INVALID_OPERATION)
763 		goto out_c;
764 
765 no_inline_bulk_status:
766 	/* We could try FS.BulkStatus next, but this aborts the entire op if
767 	 * any of the lookups fails - so, for the moment, revert to
768 	 * FS.FetchStatus for just the primary fid.
769 	 */
770 	inode = ERR_PTR(-ERESTARTSYS);
771 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
772 		while (afs_select_fileserver(&fc)) {
773 			iget_data.cb_v_break = dvnode->volume->cb_v_break;
774 			iget_data.cb_s_break = fc.cbi->server->cb_s_break;
775 			scb = &cookie->statuses[0];
776 			afs_fs_fetch_status(&fc,
777 					    afs_v2net(dvnode),
778 					    cookie->fids,
779 					    scb,
780 					    NULL);
781 		}
782 
783 		if (fc.ac.error == 0)
784 			cbi = afs_get_cb_interest(fc.cbi);
785 		inode = ERR_PTR(afs_end_vnode_operation(&fc));
786 	}
787 
788 	if (IS_ERR(inode))
789 		goto out_c;
790 
791 success:
792 	/* Turn all the files into inodes and save the first one - which is the
793 	 * one we actually want.
794 	 */
795 	scb = &cookie->statuses[0];
796 	if (scb->status.abort_code != 0)
797 		inode = ERR_PTR(afs_abort_to_error(scb->status.abort_code));
798 
799 	for (i = 0; i < cookie->nr_fids; i++) {
800 		struct afs_status_cb *scb = &cookie->statuses[i];
801 
802 		if (!scb->have_status && !scb->have_error)
803 			continue;
804 
805 		if (cookie->inodes[i]) {
806 			struct afs_vnode *iv = AFS_FS_I(cookie->inodes[i]);
807 
808 			if (test_bit(AFS_VNODE_UNSET, &iv->flags))
809 				continue;
810 
811 			afs_vnode_commit_status(&fc, iv,
812 						scb->cb_break, NULL, scb);
813 			continue;
814 		}
815 
816 		if (scb->status.abort_code != 0)
817 			continue;
818 
819 		iget_data.fid = cookie->fids[i];
820 		ti = afs_iget(dir->i_sb, key, &iget_data, scb, cbi, dvnode);
821 		if (!IS_ERR(ti))
822 			afs_cache_permit(AFS_FS_I(ti), key,
823 					 0 /* Assume vnode->cb_break is 0 */ +
824 					 iget_data.cb_v_break,
825 					 scb);
826 		if (i == 0) {
827 			inode = ti;
828 		} else {
829 			if (!IS_ERR(ti))
830 				iput(ti);
831 		}
832 	}
833 
834 out_c:
835 	afs_put_cb_interest(afs_v2net(dvnode), cbi);
836 	if (cookie->inodes) {
837 		for (i = 0; i < cookie->nr_fids; i++)
838 			iput(cookie->inodes[i]);
839 		kfree(cookie->inodes);
840 	}
841 out_s:
842 	kvfree(cookie->statuses);
843 out:
844 	kfree(cookie);
845 	return inode;
846 }
847 
848 /*
849  * Look up an entry in a directory with @sys substitution.
850  */
851 static struct dentry *afs_lookup_atsys(struct inode *dir, struct dentry *dentry,
852 				       struct key *key)
853 {
854 	struct afs_sysnames *subs;
855 	struct afs_net *net = afs_i2net(dir);
856 	struct dentry *ret;
857 	char *buf, *p, *name;
858 	int len, i;
859 
860 	_enter("");
861 
862 	ret = ERR_PTR(-ENOMEM);
863 	p = buf = kmalloc(AFSNAMEMAX, GFP_KERNEL);
864 	if (!buf)
865 		goto out_p;
866 	if (dentry->d_name.len > 4) {
867 		memcpy(p, dentry->d_name.name, dentry->d_name.len - 4);
868 		p += dentry->d_name.len - 4;
869 	}
870 
871 	/* There is an ordered list of substitutes that we have to try. */
872 	read_lock(&net->sysnames_lock);
873 	subs = net->sysnames;
874 	refcount_inc(&subs->usage);
875 	read_unlock(&net->sysnames_lock);
876 
877 	for (i = 0; i < subs->nr; i++) {
878 		name = subs->subs[i];
879 		len = dentry->d_name.len - 4 + strlen(name);
880 		if (len >= AFSNAMEMAX) {
881 			ret = ERR_PTR(-ENAMETOOLONG);
882 			goto out_s;
883 		}
884 
885 		strcpy(p, name);
886 		ret = lookup_one_len(buf, dentry->d_parent, len);
887 		if (IS_ERR(ret) || d_is_positive(ret))
888 			goto out_s;
889 		dput(ret);
890 	}
891 
892 	/* We don't want to d_add() the @sys dentry here as we don't want to
893 	 * the cached dentry to hide changes to the sysnames list.
894 	 */
895 	ret = NULL;
896 out_s:
897 	afs_put_sysnames(subs);
898 	kfree(buf);
899 out_p:
900 	key_put(key);
901 	return ret;
902 }
903 
904 /*
905  * look up an entry in a directory
906  */
907 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
908 				 unsigned int flags)
909 {
910 	struct afs_vnode *dvnode = AFS_FS_I(dir);
911 	struct afs_fid fid = {};
912 	struct inode *inode;
913 	struct dentry *d;
914 	struct key *key;
915 	int ret;
916 
917 	_enter("{%llx:%llu},%p{%pd},",
918 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, dentry);
919 
920 	ASSERTCMP(d_inode(dentry), ==, NULL);
921 
922 	if (dentry->d_name.len >= AFSNAMEMAX) {
923 		_leave(" = -ENAMETOOLONG");
924 		return ERR_PTR(-ENAMETOOLONG);
925 	}
926 
927 	if (test_bit(AFS_VNODE_DELETED, &dvnode->flags)) {
928 		_leave(" = -ESTALE");
929 		return ERR_PTR(-ESTALE);
930 	}
931 
932 	key = afs_request_key(dvnode->volume->cell);
933 	if (IS_ERR(key)) {
934 		_leave(" = %ld [key]", PTR_ERR(key));
935 		return ERR_CAST(key);
936 	}
937 
938 	ret = afs_validate(dvnode, key);
939 	if (ret < 0) {
940 		key_put(key);
941 		_leave(" = %d [val]", ret);
942 		return ERR_PTR(ret);
943 	}
944 
945 	if (dentry->d_name.len >= 4 &&
946 	    dentry->d_name.name[dentry->d_name.len - 4] == '@' &&
947 	    dentry->d_name.name[dentry->d_name.len - 3] == 's' &&
948 	    dentry->d_name.name[dentry->d_name.len - 2] == 'y' &&
949 	    dentry->d_name.name[dentry->d_name.len - 1] == 's')
950 		return afs_lookup_atsys(dir, dentry, key);
951 
952 	afs_stat_v(dvnode, n_lookup);
953 	inode = afs_do_lookup(dir, dentry, key);
954 	key_put(key);
955 	if (inode == ERR_PTR(-ENOENT))
956 		inode = afs_try_auto_mntpt(dentry, dir);
957 
958 	if (!IS_ERR_OR_NULL(inode))
959 		fid = AFS_FS_I(inode)->fid;
960 
961 	d = d_splice_alias(inode, dentry);
962 	if (!IS_ERR_OR_NULL(d)) {
963 		d->d_fsdata = dentry->d_fsdata;
964 		trace_afs_lookup(dvnode, &d->d_name, &fid);
965 	} else {
966 		trace_afs_lookup(dvnode, &dentry->d_name, &fid);
967 	}
968 	return d;
969 }
970 
971 /*
972  * Check the validity of a dentry under RCU conditions.
973  */
974 static int afs_d_revalidate_rcu(struct dentry *dentry)
975 {
976 	struct afs_vnode *dvnode, *vnode;
977 	struct dentry *parent;
978 	struct inode *dir, *inode;
979 	long dir_version, de_version;
980 
981 	_enter("%p", dentry);
982 
983 	/* Check the parent directory is still valid first. */
984 	parent = READ_ONCE(dentry->d_parent);
985 	dir = d_inode_rcu(parent);
986 	if (!dir)
987 		return -ECHILD;
988 	dvnode = AFS_FS_I(dir);
989 	if (test_bit(AFS_VNODE_DELETED, &dvnode->flags))
990 		return -ECHILD;
991 
992 	if (!afs_check_validity(dvnode))
993 		return -ECHILD;
994 
995 	/* We only need to invalidate a dentry if the server's copy changed
996 	 * behind our back.  If we made the change, it's no problem.  Note that
997 	 * on a 32-bit system, we only have 32 bits in the dentry to store the
998 	 * version.
999 	 */
1000 	dir_version = (long)READ_ONCE(dvnode->status.data_version);
1001 	de_version = (long)READ_ONCE(dentry->d_fsdata);
1002 	if (de_version != dir_version) {
1003 		dir_version = (long)READ_ONCE(dvnode->invalid_before);
1004 		if (de_version - dir_version < 0)
1005 			return -ECHILD;
1006 	}
1007 
1008 	/* Check to see if the vnode referred to by the dentry still
1009 	 * has a callback.
1010 	 */
1011 	if (d_really_is_positive(dentry)) {
1012 		inode = d_inode_rcu(dentry);
1013 		if (inode) {
1014 			vnode = AFS_FS_I(inode);
1015 			if (!afs_check_validity(vnode))
1016 				return -ECHILD;
1017 		}
1018 	}
1019 
1020 	return 1; /* Still valid */
1021 }
1022 
1023 /*
1024  * check that a dentry lookup hit has found a valid entry
1025  * - NOTE! the hit can be a negative hit too, so we can't assume we have an
1026  *   inode
1027  */
1028 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags)
1029 {
1030 	struct afs_vnode *vnode, *dir;
1031 	struct afs_fid uninitialized_var(fid);
1032 	struct dentry *parent;
1033 	struct inode *inode;
1034 	struct key *key;
1035 	afs_dataversion_t dir_version;
1036 	long de_version;
1037 	int ret;
1038 
1039 	if (flags & LOOKUP_RCU)
1040 		return afs_d_revalidate_rcu(dentry);
1041 
1042 	if (d_really_is_positive(dentry)) {
1043 		vnode = AFS_FS_I(d_inode(dentry));
1044 		_enter("{v={%llx:%llu} n=%pd fl=%lx},",
1045 		       vnode->fid.vid, vnode->fid.vnode, dentry,
1046 		       vnode->flags);
1047 	} else {
1048 		_enter("{neg n=%pd}", dentry);
1049 	}
1050 
1051 	key = afs_request_key(AFS_FS_S(dentry->d_sb)->volume->cell);
1052 	if (IS_ERR(key))
1053 		key = NULL;
1054 
1055 	if (d_really_is_positive(dentry)) {
1056 		inode = d_inode(dentry);
1057 		if (inode) {
1058 			vnode = AFS_FS_I(inode);
1059 			afs_validate(vnode, key);
1060 			if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1061 				goto out_bad;
1062 		}
1063 	}
1064 
1065 	/* lock down the parent dentry so we can peer at it */
1066 	parent = dget_parent(dentry);
1067 	dir = AFS_FS_I(d_inode(parent));
1068 
1069 	/* validate the parent directory */
1070 	afs_validate(dir, key);
1071 
1072 	if (test_bit(AFS_VNODE_DELETED, &dir->flags)) {
1073 		_debug("%pd: parent dir deleted", dentry);
1074 		goto out_bad_parent;
1075 	}
1076 
1077 	/* We only need to invalidate a dentry if the server's copy changed
1078 	 * behind our back.  If we made the change, it's no problem.  Note that
1079 	 * on a 32-bit system, we only have 32 bits in the dentry to store the
1080 	 * version.
1081 	 */
1082 	dir_version = dir->status.data_version;
1083 	de_version = (long)dentry->d_fsdata;
1084 	if (de_version == (long)dir_version)
1085 		goto out_valid_noupdate;
1086 
1087 	dir_version = dir->invalid_before;
1088 	if (de_version - (long)dir_version >= 0)
1089 		goto out_valid;
1090 
1091 	_debug("dir modified");
1092 	afs_stat_v(dir, n_reval);
1093 
1094 	/* search the directory for this vnode */
1095 	ret = afs_do_lookup_one(&dir->vfs_inode, dentry, &fid, key, &dir_version);
1096 	switch (ret) {
1097 	case 0:
1098 		/* the filename maps to something */
1099 		if (d_really_is_negative(dentry))
1100 			goto out_bad_parent;
1101 		inode = d_inode(dentry);
1102 		if (is_bad_inode(inode)) {
1103 			printk("kAFS: afs_d_revalidate: %pd2 has bad inode\n",
1104 			       dentry);
1105 			goto out_bad_parent;
1106 		}
1107 
1108 		vnode = AFS_FS_I(inode);
1109 
1110 		/* if the vnode ID has changed, then the dirent points to a
1111 		 * different file */
1112 		if (fid.vnode != vnode->fid.vnode) {
1113 			_debug("%pd: dirent changed [%llu != %llu]",
1114 			       dentry, fid.vnode,
1115 			       vnode->fid.vnode);
1116 			goto not_found;
1117 		}
1118 
1119 		/* if the vnode ID uniqifier has changed, then the file has
1120 		 * been deleted and replaced, and the original vnode ID has
1121 		 * been reused */
1122 		if (fid.unique != vnode->fid.unique) {
1123 			_debug("%pd: file deleted (uq %u -> %u I:%u)",
1124 			       dentry, fid.unique,
1125 			       vnode->fid.unique,
1126 			       vnode->vfs_inode.i_generation);
1127 			write_seqlock(&vnode->cb_lock);
1128 			set_bit(AFS_VNODE_DELETED, &vnode->flags);
1129 			write_sequnlock(&vnode->cb_lock);
1130 			goto not_found;
1131 		}
1132 		goto out_valid;
1133 
1134 	case -ENOENT:
1135 		/* the filename is unknown */
1136 		_debug("%pd: dirent not found", dentry);
1137 		if (d_really_is_positive(dentry))
1138 			goto not_found;
1139 		goto out_valid;
1140 
1141 	default:
1142 		_debug("failed to iterate dir %pd: %d",
1143 		       parent, ret);
1144 		goto out_bad_parent;
1145 	}
1146 
1147 out_valid:
1148 	dentry->d_fsdata = (void *)(unsigned long)dir_version;
1149 out_valid_noupdate:
1150 	dput(parent);
1151 	key_put(key);
1152 	_leave(" = 1 [valid]");
1153 	return 1;
1154 
1155 	/* the dirent, if it exists, now points to a different vnode */
1156 not_found:
1157 	spin_lock(&dentry->d_lock);
1158 	dentry->d_flags |= DCACHE_NFSFS_RENAMED;
1159 	spin_unlock(&dentry->d_lock);
1160 
1161 out_bad_parent:
1162 	_debug("dropping dentry %pd2", dentry);
1163 	dput(parent);
1164 out_bad:
1165 	key_put(key);
1166 
1167 	_leave(" = 0 [bad]");
1168 	return 0;
1169 }
1170 
1171 /*
1172  * allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
1173  * sleep)
1174  * - called from dput() when d_count is going to 0.
1175  * - return 1 to request dentry be unhashed, 0 otherwise
1176  */
1177 static int afs_d_delete(const struct dentry *dentry)
1178 {
1179 	_enter("%pd", dentry);
1180 
1181 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1182 		goto zap;
1183 
1184 	if (d_really_is_positive(dentry) &&
1185 	    (test_bit(AFS_VNODE_DELETED,   &AFS_FS_I(d_inode(dentry))->flags) ||
1186 	     test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(d_inode(dentry))->flags)))
1187 		goto zap;
1188 
1189 	_leave(" = 0 [keep]");
1190 	return 0;
1191 
1192 zap:
1193 	_leave(" = 1 [zap]");
1194 	return 1;
1195 }
1196 
1197 /*
1198  * Clean up sillyrename files on dentry removal.
1199  */
1200 static void afs_d_iput(struct dentry *dentry, struct inode *inode)
1201 {
1202 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1203 		afs_silly_iput(dentry, inode);
1204 	iput(inode);
1205 }
1206 
1207 /*
1208  * handle dentry release
1209  */
1210 void afs_d_release(struct dentry *dentry)
1211 {
1212 	_enter("%pd", dentry);
1213 }
1214 
1215 /*
1216  * Create a new inode for create/mkdir/symlink
1217  */
1218 static void afs_vnode_new_inode(struct afs_fs_cursor *fc,
1219 				struct dentry *new_dentry,
1220 				struct afs_iget_data *new_data,
1221 				struct afs_status_cb *new_scb)
1222 {
1223 	struct afs_vnode *vnode;
1224 	struct inode *inode;
1225 
1226 	if (fc->ac.error < 0)
1227 		return;
1228 
1229 	inode = afs_iget(fc->vnode->vfs_inode.i_sb, fc->key,
1230 			 new_data, new_scb, fc->cbi, fc->vnode);
1231 	if (IS_ERR(inode)) {
1232 		/* ENOMEM or EINTR at a really inconvenient time - just abandon
1233 		 * the new directory on the server.
1234 		 */
1235 		fc->ac.error = PTR_ERR(inode);
1236 		return;
1237 	}
1238 
1239 	vnode = AFS_FS_I(inode);
1240 	set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
1241 	if (fc->ac.error == 0)
1242 		afs_cache_permit(vnode, fc->key, vnode->cb_break, new_scb);
1243 	d_instantiate(new_dentry, inode);
1244 }
1245 
1246 static void afs_prep_for_new_inode(struct afs_fs_cursor *fc,
1247 				   struct afs_iget_data *iget_data)
1248 {
1249 	iget_data->volume = fc->vnode->volume;
1250 	iget_data->cb_v_break = fc->vnode->volume->cb_v_break;
1251 	iget_data->cb_s_break = fc->cbi->server->cb_s_break;
1252 }
1253 
1254 /*
1255  * Note that a dentry got changed.  We need to set d_fsdata to the data version
1256  * number derived from the result of the operation.  It doesn't matter if
1257  * d_fsdata goes backwards as we'll just revalidate.
1258  */
1259 static void afs_update_dentry_version(struct afs_fs_cursor *fc,
1260 				      struct dentry *dentry,
1261 				      struct afs_status_cb *scb)
1262 {
1263 	if (fc->ac.error == 0)
1264 		dentry->d_fsdata =
1265 			(void *)(unsigned long)scb->status.data_version;
1266 }
1267 
1268 /*
1269  * create a directory on an AFS filesystem
1270  */
1271 static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1272 {
1273 	struct afs_iget_data iget_data;
1274 	struct afs_status_cb *scb;
1275 	struct afs_fs_cursor fc;
1276 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1277 	struct key *key;
1278 	int ret;
1279 
1280 	mode |= S_IFDIR;
1281 
1282 	_enter("{%llx:%llu},{%pd},%ho",
1283 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1284 
1285 	ret = -ENOMEM;
1286 	scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1287 	if (!scb)
1288 		goto error;
1289 
1290 	key = afs_request_key(dvnode->volume->cell);
1291 	if (IS_ERR(key)) {
1292 		ret = PTR_ERR(key);
1293 		goto error_scb;
1294 	}
1295 
1296 	ret = -ERESTARTSYS;
1297 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1298 		afs_dataversion_t data_version = dvnode->status.data_version + 1;
1299 
1300 		while (afs_select_fileserver(&fc)) {
1301 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1302 			afs_prep_for_new_inode(&fc, &iget_data);
1303 			afs_fs_create(&fc, dentry->d_name.name, mode,
1304 				      &scb[0], &iget_data.fid, &scb[1]);
1305 		}
1306 
1307 		afs_check_for_remote_deletion(&fc, dvnode);
1308 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1309 					&data_version, &scb[0]);
1310 		afs_update_dentry_version(&fc, dentry, &scb[0]);
1311 		afs_vnode_new_inode(&fc, dentry, &iget_data, &scb[1]);
1312 		ret = afs_end_vnode_operation(&fc);
1313 		if (ret < 0)
1314 			goto error_key;
1315 	} else {
1316 		goto error_key;
1317 	}
1318 
1319 	if (ret == 0 &&
1320 	    test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1321 		afs_edit_dir_add(dvnode, &dentry->d_name, &iget_data.fid,
1322 				 afs_edit_dir_for_create);
1323 
1324 	key_put(key);
1325 	kfree(scb);
1326 	_leave(" = 0");
1327 	return 0;
1328 
1329 error_key:
1330 	key_put(key);
1331 error_scb:
1332 	kfree(scb);
1333 error:
1334 	d_drop(dentry);
1335 	_leave(" = %d", ret);
1336 	return ret;
1337 }
1338 
1339 /*
1340  * Remove a subdir from a directory.
1341  */
1342 static void afs_dir_remove_subdir(struct dentry *dentry)
1343 {
1344 	if (d_really_is_positive(dentry)) {
1345 		struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1346 
1347 		clear_nlink(&vnode->vfs_inode);
1348 		set_bit(AFS_VNODE_DELETED, &vnode->flags);
1349 		clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1350 		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
1351 	}
1352 }
1353 
1354 /*
1355  * remove a directory from an AFS filesystem
1356  */
1357 static int afs_rmdir(struct inode *dir, struct dentry *dentry)
1358 {
1359 	struct afs_status_cb *scb;
1360 	struct afs_fs_cursor fc;
1361 	struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode = NULL;
1362 	struct key *key;
1363 	int ret;
1364 
1365 	_enter("{%llx:%llu},{%pd}",
1366 	       dvnode->fid.vid, dvnode->fid.vnode, dentry);
1367 
1368 	scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL);
1369 	if (!scb)
1370 		return -ENOMEM;
1371 
1372 	key = afs_request_key(dvnode->volume->cell);
1373 	if (IS_ERR(key)) {
1374 		ret = PTR_ERR(key);
1375 		goto error;
1376 	}
1377 
1378 	/* Try to make sure we have a callback promise on the victim. */
1379 	if (d_really_is_positive(dentry)) {
1380 		vnode = AFS_FS_I(d_inode(dentry));
1381 		ret = afs_validate(vnode, key);
1382 		if (ret < 0)
1383 			goto error_key;
1384 	}
1385 
1386 	if (vnode) {
1387 		ret = down_write_killable(&vnode->rmdir_lock);
1388 		if (ret < 0)
1389 			goto error_key;
1390 	}
1391 
1392 	ret = -ERESTARTSYS;
1393 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1394 		afs_dataversion_t data_version = dvnode->status.data_version + 1;
1395 
1396 		while (afs_select_fileserver(&fc)) {
1397 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1398 			afs_fs_remove(&fc, vnode, dentry->d_name.name, true, scb);
1399 		}
1400 
1401 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1402 					&data_version, scb);
1403 		afs_update_dentry_version(&fc, dentry, scb);
1404 		ret = afs_end_vnode_operation(&fc);
1405 		if (ret == 0) {
1406 			afs_dir_remove_subdir(dentry);
1407 			if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1408 				afs_edit_dir_remove(dvnode, &dentry->d_name,
1409 						    afs_edit_dir_for_rmdir);
1410 		}
1411 	}
1412 
1413 	if (vnode)
1414 		up_write(&vnode->rmdir_lock);
1415 error_key:
1416 	key_put(key);
1417 error:
1418 	kfree(scb);
1419 	return ret;
1420 }
1421 
1422 /*
1423  * Remove a link to a file or symlink from a directory.
1424  *
1425  * If the file was not deleted due to excess hard links, the fileserver will
1426  * break the callback promise on the file - if it had one - before it returns
1427  * to us, and if it was deleted, it won't
1428  *
1429  * However, if we didn't have a callback promise outstanding, or it was
1430  * outstanding on a different server, then it won't break it either...
1431  */
1432 static int afs_dir_remove_link(struct afs_vnode *dvnode, struct dentry *dentry,
1433 			       struct key *key)
1434 {
1435 	int ret = 0;
1436 
1437 	if (d_really_is_positive(dentry)) {
1438 		struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1439 
1440 		if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
1441 			/* Already done */
1442 		} else if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
1443 			write_seqlock(&vnode->cb_lock);
1444 			drop_nlink(&vnode->vfs_inode);
1445 			if (vnode->vfs_inode.i_nlink == 0) {
1446 				set_bit(AFS_VNODE_DELETED, &vnode->flags);
1447 				__afs_break_callback(vnode, afs_cb_break_for_unlink);
1448 			}
1449 			write_sequnlock(&vnode->cb_lock);
1450 			ret = 0;
1451 		} else {
1452 			afs_break_callback(vnode, afs_cb_break_for_unlink);
1453 
1454 			if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1455 				kdebug("AFS_VNODE_DELETED");
1456 
1457 			ret = afs_validate(vnode, key);
1458 			if (ret == -ESTALE)
1459 				ret = 0;
1460 		}
1461 		_debug("nlink %d [val %d]", vnode->vfs_inode.i_nlink, ret);
1462 	}
1463 
1464 	return ret;
1465 }
1466 
1467 /*
1468  * Remove a file or symlink from an AFS filesystem.
1469  */
1470 static int afs_unlink(struct inode *dir, struct dentry *dentry)
1471 {
1472 	struct afs_fs_cursor fc;
1473 	struct afs_status_cb *scb;
1474 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1475 	struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1476 	struct key *key;
1477 	bool need_rehash = false;
1478 	int ret;
1479 
1480 	_enter("{%llx:%llu},{%pd}",
1481 	       dvnode->fid.vid, dvnode->fid.vnode, dentry);
1482 
1483 	if (dentry->d_name.len >= AFSNAMEMAX)
1484 		return -ENAMETOOLONG;
1485 
1486 	ret = -ENOMEM;
1487 	scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1488 	if (!scb)
1489 		goto error;
1490 
1491 	key = afs_request_key(dvnode->volume->cell);
1492 	if (IS_ERR(key)) {
1493 		ret = PTR_ERR(key);
1494 		goto error_scb;
1495 	}
1496 
1497 	/* Try to make sure we have a callback promise on the victim. */
1498 	ret = afs_validate(vnode, key);
1499 	if (ret < 0)
1500 		goto error_key;
1501 
1502 	spin_lock(&dentry->d_lock);
1503 	if (d_count(dentry) > 1) {
1504 		spin_unlock(&dentry->d_lock);
1505 		/* Start asynchronous writeout of the inode */
1506 		write_inode_now(d_inode(dentry), 0);
1507 		ret = afs_sillyrename(dvnode, vnode, dentry, key);
1508 		goto error_key;
1509 	}
1510 	if (!d_unhashed(dentry)) {
1511 		/* Prevent a race with RCU lookup. */
1512 		__d_drop(dentry);
1513 		need_rehash = true;
1514 	}
1515 	spin_unlock(&dentry->d_lock);
1516 
1517 	ret = -ERESTARTSYS;
1518 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1519 		afs_dataversion_t data_version = dvnode->status.data_version + 1;
1520 		afs_dataversion_t data_version_2 = vnode->status.data_version;
1521 
1522 		while (afs_select_fileserver(&fc)) {
1523 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1524 			fc.cb_break_2 = afs_calc_vnode_cb_break(vnode);
1525 
1526 			if (test_bit(AFS_SERVER_FL_IS_YFS, &fc.cbi->server->flags) &&
1527 			    !test_bit(AFS_SERVER_FL_NO_RM2, &fc.cbi->server->flags)) {
1528 				yfs_fs_remove_file2(&fc, vnode, dentry->d_name.name,
1529 						    &scb[0], &scb[1]);
1530 				if (fc.ac.error != -ECONNABORTED ||
1531 				    fc.ac.abort_code != RXGEN_OPCODE)
1532 					continue;
1533 				set_bit(AFS_SERVER_FL_NO_RM2, &fc.cbi->server->flags);
1534 			}
1535 
1536 			afs_fs_remove(&fc, vnode, dentry->d_name.name, false, &scb[0]);
1537 		}
1538 
1539 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1540 					&data_version, &scb[0]);
1541 		afs_vnode_commit_status(&fc, vnode, fc.cb_break_2,
1542 					&data_version_2, &scb[1]);
1543 		afs_update_dentry_version(&fc, dentry, &scb[0]);
1544 		ret = afs_end_vnode_operation(&fc);
1545 		if (ret == 0 && !(scb[1].have_status || scb[1].have_error))
1546 			ret = afs_dir_remove_link(dvnode, dentry, key);
1547 		if (ret == 0 &&
1548 		    test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1549 			afs_edit_dir_remove(dvnode, &dentry->d_name,
1550 					    afs_edit_dir_for_unlink);
1551 	}
1552 
1553 	if (need_rehash && ret < 0 && ret != -ENOENT)
1554 		d_rehash(dentry);
1555 
1556 error_key:
1557 	key_put(key);
1558 error_scb:
1559 	kfree(scb);
1560 error:
1561 	_leave(" = %d", ret);
1562 	return ret;
1563 }
1564 
1565 /*
1566  * create a regular file on an AFS filesystem
1567  */
1568 static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1569 		      bool excl)
1570 {
1571 	struct afs_iget_data iget_data;
1572 	struct afs_fs_cursor fc;
1573 	struct afs_status_cb *scb;
1574 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1575 	struct key *key;
1576 	int ret;
1577 
1578 	mode |= S_IFREG;
1579 
1580 	_enter("{%llx:%llu},{%pd},%ho,",
1581 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1582 
1583 	ret = -ENAMETOOLONG;
1584 	if (dentry->d_name.len >= AFSNAMEMAX)
1585 		goto error;
1586 
1587 	key = afs_request_key(dvnode->volume->cell);
1588 	if (IS_ERR(key)) {
1589 		ret = PTR_ERR(key);
1590 		goto error;
1591 	}
1592 
1593 	ret = -ENOMEM;
1594 	scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1595 	if (!scb)
1596 		goto error_scb;
1597 
1598 	ret = -ERESTARTSYS;
1599 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1600 		afs_dataversion_t data_version = dvnode->status.data_version + 1;
1601 
1602 		while (afs_select_fileserver(&fc)) {
1603 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1604 			afs_prep_for_new_inode(&fc, &iget_data);
1605 			afs_fs_create(&fc, dentry->d_name.name, mode,
1606 				      &scb[0], &iget_data.fid, &scb[1]);
1607 		}
1608 
1609 		afs_check_for_remote_deletion(&fc, dvnode);
1610 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1611 					&data_version, &scb[0]);
1612 		afs_update_dentry_version(&fc, dentry, &scb[0]);
1613 		afs_vnode_new_inode(&fc, dentry, &iget_data, &scb[1]);
1614 		ret = afs_end_vnode_operation(&fc);
1615 		if (ret < 0)
1616 			goto error_key;
1617 	} else {
1618 		goto error_key;
1619 	}
1620 
1621 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1622 		afs_edit_dir_add(dvnode, &dentry->d_name, &iget_data.fid,
1623 				 afs_edit_dir_for_create);
1624 
1625 	kfree(scb);
1626 	key_put(key);
1627 	_leave(" = 0");
1628 	return 0;
1629 
1630 error_scb:
1631 	kfree(scb);
1632 error_key:
1633 	key_put(key);
1634 error:
1635 	d_drop(dentry);
1636 	_leave(" = %d", ret);
1637 	return ret;
1638 }
1639 
1640 /*
1641  * create a hard link between files in an AFS filesystem
1642  */
1643 static int afs_link(struct dentry *from, struct inode *dir,
1644 		    struct dentry *dentry)
1645 {
1646 	struct afs_fs_cursor fc;
1647 	struct afs_status_cb *scb;
1648 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1649 	struct afs_vnode *vnode = AFS_FS_I(d_inode(from));
1650 	struct key *key;
1651 	int ret;
1652 
1653 	_enter("{%llx:%llu},{%llx:%llu},{%pd}",
1654 	       vnode->fid.vid, vnode->fid.vnode,
1655 	       dvnode->fid.vid, dvnode->fid.vnode,
1656 	       dentry);
1657 
1658 	ret = -ENAMETOOLONG;
1659 	if (dentry->d_name.len >= AFSNAMEMAX)
1660 		goto error;
1661 
1662 	ret = -ENOMEM;
1663 	scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1664 	if (!scb)
1665 		goto error;
1666 
1667 	key = afs_request_key(dvnode->volume->cell);
1668 	if (IS_ERR(key)) {
1669 		ret = PTR_ERR(key);
1670 		goto error_scb;
1671 	}
1672 
1673 	ret = -ERESTARTSYS;
1674 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1675 		afs_dataversion_t data_version = dvnode->status.data_version + 1;
1676 
1677 		if (mutex_lock_interruptible_nested(&vnode->io_lock, 1) < 0) {
1678 			afs_end_vnode_operation(&fc);
1679 			goto error_key;
1680 		}
1681 
1682 		while (afs_select_fileserver(&fc)) {
1683 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1684 			fc.cb_break_2 = afs_calc_vnode_cb_break(vnode);
1685 			afs_fs_link(&fc, vnode, dentry->d_name.name,
1686 				    &scb[0], &scb[1]);
1687 		}
1688 
1689 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1690 					&data_version, &scb[0]);
1691 		afs_vnode_commit_status(&fc, vnode, fc.cb_break_2,
1692 					NULL, &scb[1]);
1693 		ihold(&vnode->vfs_inode);
1694 		afs_update_dentry_version(&fc, dentry, &scb[0]);
1695 		d_instantiate(dentry, &vnode->vfs_inode);
1696 
1697 		mutex_unlock(&vnode->io_lock);
1698 		ret = afs_end_vnode_operation(&fc);
1699 		if (ret < 0)
1700 			goto error_key;
1701 	} else {
1702 		goto error_key;
1703 	}
1704 
1705 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1706 		afs_edit_dir_add(dvnode, &dentry->d_name, &vnode->fid,
1707 				 afs_edit_dir_for_link);
1708 
1709 	key_put(key);
1710 	kfree(scb);
1711 	_leave(" = 0");
1712 	return 0;
1713 
1714 error_key:
1715 	key_put(key);
1716 error_scb:
1717 	kfree(scb);
1718 error:
1719 	d_drop(dentry);
1720 	_leave(" = %d", ret);
1721 	return ret;
1722 }
1723 
1724 /*
1725  * create a symlink in an AFS filesystem
1726  */
1727 static int afs_symlink(struct inode *dir, struct dentry *dentry,
1728 		       const char *content)
1729 {
1730 	struct afs_iget_data iget_data;
1731 	struct afs_fs_cursor fc;
1732 	struct afs_status_cb *scb;
1733 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1734 	struct key *key;
1735 	int ret;
1736 
1737 	_enter("{%llx:%llu},{%pd},%s",
1738 	       dvnode->fid.vid, dvnode->fid.vnode, dentry,
1739 	       content);
1740 
1741 	ret = -ENAMETOOLONG;
1742 	if (dentry->d_name.len >= AFSNAMEMAX)
1743 		goto error;
1744 
1745 	ret = -EINVAL;
1746 	if (strlen(content) >= AFSPATHMAX)
1747 		goto error;
1748 
1749 	ret = -ENOMEM;
1750 	scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1751 	if (!scb)
1752 		goto error;
1753 
1754 	key = afs_request_key(dvnode->volume->cell);
1755 	if (IS_ERR(key)) {
1756 		ret = PTR_ERR(key);
1757 		goto error_scb;
1758 	}
1759 
1760 	ret = -ERESTARTSYS;
1761 	if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1762 		afs_dataversion_t data_version = dvnode->status.data_version + 1;
1763 
1764 		while (afs_select_fileserver(&fc)) {
1765 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1766 			afs_prep_for_new_inode(&fc, &iget_data);
1767 			afs_fs_symlink(&fc, dentry->d_name.name, content,
1768 				       &scb[0], &iget_data.fid, &scb[1]);
1769 		}
1770 
1771 		afs_check_for_remote_deletion(&fc, dvnode);
1772 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1773 					&data_version, &scb[0]);
1774 		afs_update_dentry_version(&fc, dentry, &scb[0]);
1775 		afs_vnode_new_inode(&fc, dentry, &iget_data, &scb[1]);
1776 		ret = afs_end_vnode_operation(&fc);
1777 		if (ret < 0)
1778 			goto error_key;
1779 	} else {
1780 		goto error_key;
1781 	}
1782 
1783 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1784 		afs_edit_dir_add(dvnode, &dentry->d_name, &iget_data.fid,
1785 				 afs_edit_dir_for_symlink);
1786 
1787 	key_put(key);
1788 	kfree(scb);
1789 	_leave(" = 0");
1790 	return 0;
1791 
1792 error_key:
1793 	key_put(key);
1794 error_scb:
1795 	kfree(scb);
1796 error:
1797 	d_drop(dentry);
1798 	_leave(" = %d", ret);
1799 	return ret;
1800 }
1801 
1802 /*
1803  * rename a file in an AFS filesystem and/or move it between directories
1804  */
1805 static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
1806 		      struct inode *new_dir, struct dentry *new_dentry,
1807 		      unsigned int flags)
1808 {
1809 	struct afs_fs_cursor fc;
1810 	struct afs_status_cb *scb;
1811 	struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
1812 	struct dentry *tmp = NULL, *rehash = NULL;
1813 	struct inode *new_inode;
1814 	struct key *key;
1815 	bool new_negative = d_is_negative(new_dentry);
1816 	int ret;
1817 
1818 	if (flags)
1819 		return -EINVAL;
1820 
1821 	/* Don't allow silly-rename files be moved around. */
1822 	if (old_dentry->d_flags & DCACHE_NFSFS_RENAMED)
1823 		return -EINVAL;
1824 
1825 	vnode = AFS_FS_I(d_inode(old_dentry));
1826 	orig_dvnode = AFS_FS_I(old_dir);
1827 	new_dvnode = AFS_FS_I(new_dir);
1828 
1829 	_enter("{%llx:%llu},{%llx:%llu},{%llx:%llu},{%pd}",
1830 	       orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
1831 	       vnode->fid.vid, vnode->fid.vnode,
1832 	       new_dvnode->fid.vid, new_dvnode->fid.vnode,
1833 	       new_dentry);
1834 
1835 	ret = -ENOMEM;
1836 	scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1837 	if (!scb)
1838 		goto error;
1839 
1840 	key = afs_request_key(orig_dvnode->volume->cell);
1841 	if (IS_ERR(key)) {
1842 		ret = PTR_ERR(key);
1843 		goto error_scb;
1844 	}
1845 
1846 	/* For non-directories, check whether the target is busy and if so,
1847 	 * make a copy of the dentry and then do a silly-rename.  If the
1848 	 * silly-rename succeeds, the copied dentry is hashed and becomes the
1849 	 * new target.
1850 	 */
1851 	if (d_is_positive(new_dentry) && !d_is_dir(new_dentry)) {
1852 		/* To prevent any new references to the target during the
1853 		 * rename, we unhash the dentry in advance.
1854 		 */
1855 		if (!d_unhashed(new_dentry)) {
1856 			d_drop(new_dentry);
1857 			rehash = new_dentry;
1858 		}
1859 
1860 		if (d_count(new_dentry) > 2) {
1861 			/* copy the target dentry's name */
1862 			ret = -ENOMEM;
1863 			tmp = d_alloc(new_dentry->d_parent,
1864 				      &new_dentry->d_name);
1865 			if (!tmp)
1866 				goto error_rehash;
1867 
1868 			ret = afs_sillyrename(new_dvnode,
1869 					      AFS_FS_I(d_inode(new_dentry)),
1870 					      new_dentry, key);
1871 			if (ret)
1872 				goto error_rehash;
1873 
1874 			new_dentry = tmp;
1875 			rehash = NULL;
1876 			new_negative = true;
1877 		}
1878 	}
1879 
1880 	/* This bit is potentially nasty as there's a potential race with
1881 	 * afs_d_revalidate{,_rcu}().  We have to change d_fsdata on the dentry
1882 	 * to reflect it's new parent's new data_version after the op, but
1883 	 * d_revalidate may see old_dentry between the op having taken place
1884 	 * and the version being updated.
1885 	 *
1886 	 * So drop the old_dentry for now to make other threads go through
1887 	 * lookup instead - which we hold a lock against.
1888 	 */
1889 	d_drop(old_dentry);
1890 
1891 	ret = -ERESTARTSYS;
1892 	if (afs_begin_vnode_operation(&fc, orig_dvnode, key, true)) {
1893 		afs_dataversion_t orig_data_version;
1894 		afs_dataversion_t new_data_version;
1895 		struct afs_status_cb *new_scb = &scb[1];
1896 
1897 		orig_data_version = orig_dvnode->status.data_version + 1;
1898 
1899 		if (orig_dvnode != new_dvnode) {
1900 			if (mutex_lock_interruptible_nested(&new_dvnode->io_lock, 1) < 0) {
1901 				afs_end_vnode_operation(&fc);
1902 				goto error_rehash_old;
1903 			}
1904 			new_data_version = new_dvnode->status.data_version + 1;
1905 		} else {
1906 			new_data_version = orig_data_version;
1907 			new_scb = &scb[0];
1908 		}
1909 
1910 		while (afs_select_fileserver(&fc)) {
1911 			fc.cb_break = afs_calc_vnode_cb_break(orig_dvnode);
1912 			fc.cb_break_2 = afs_calc_vnode_cb_break(new_dvnode);
1913 			afs_fs_rename(&fc, old_dentry->d_name.name,
1914 				      new_dvnode, new_dentry->d_name.name,
1915 				      &scb[0], new_scb);
1916 		}
1917 
1918 		afs_vnode_commit_status(&fc, orig_dvnode, fc.cb_break,
1919 					&orig_data_version, &scb[0]);
1920 		if (new_dvnode != orig_dvnode) {
1921 			afs_vnode_commit_status(&fc, new_dvnode, fc.cb_break_2,
1922 						&new_data_version, &scb[1]);
1923 			mutex_unlock(&new_dvnode->io_lock);
1924 		}
1925 		ret = afs_end_vnode_operation(&fc);
1926 		if (ret < 0)
1927 			goto error_rehash_old;
1928 	}
1929 
1930 	if (ret == 0) {
1931 		if (rehash)
1932 			d_rehash(rehash);
1933 		if (test_bit(AFS_VNODE_DIR_VALID, &orig_dvnode->flags))
1934 		    afs_edit_dir_remove(orig_dvnode, &old_dentry->d_name,
1935 					afs_edit_dir_for_rename_0);
1936 
1937 		if (!new_negative &&
1938 		    test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
1939 			afs_edit_dir_remove(new_dvnode, &new_dentry->d_name,
1940 					    afs_edit_dir_for_rename_1);
1941 
1942 		if (test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
1943 			afs_edit_dir_add(new_dvnode, &new_dentry->d_name,
1944 					 &vnode->fid, afs_edit_dir_for_rename_2);
1945 
1946 		new_inode = d_inode(new_dentry);
1947 		if (new_inode) {
1948 			spin_lock(&new_inode->i_lock);
1949 			if (new_inode->i_nlink > 0)
1950 				drop_nlink(new_inode);
1951 			spin_unlock(&new_inode->i_lock);
1952 		}
1953 
1954 		/* Now we can update d_fsdata on the dentries to reflect their
1955 		 * new parent's data_version.
1956 		 *
1957 		 * Note that if we ever implement RENAME_EXCHANGE, we'll have
1958 		 * to update both dentries with opposing dir versions.
1959 		 */
1960 		if (new_dvnode != orig_dvnode) {
1961 			afs_update_dentry_version(&fc, old_dentry, &scb[1]);
1962 			afs_update_dentry_version(&fc, new_dentry, &scb[1]);
1963 		} else {
1964 			afs_update_dentry_version(&fc, old_dentry, &scb[0]);
1965 			afs_update_dentry_version(&fc, new_dentry, &scb[0]);
1966 		}
1967 		d_move(old_dentry, new_dentry);
1968 		goto error_tmp;
1969 	}
1970 
1971 error_rehash_old:
1972 	d_rehash(new_dentry);
1973 error_rehash:
1974 	if (rehash)
1975 		d_rehash(rehash);
1976 error_tmp:
1977 	if (tmp)
1978 		dput(tmp);
1979 	key_put(key);
1980 error_scb:
1981 	kfree(scb);
1982 error:
1983 	_leave(" = %d", ret);
1984 	return ret;
1985 }
1986 
1987 /*
1988  * Release a directory page and clean up its private state if it's not busy
1989  * - return true if the page can now be released, false if not
1990  */
1991 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags)
1992 {
1993 	struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
1994 
1995 	_enter("{{%llx:%llu}[%lu]}", dvnode->fid.vid, dvnode->fid.vnode, page->index);
1996 
1997 	set_page_private(page, 0);
1998 	ClearPagePrivate(page);
1999 
2000 	/* The directory will need reloading. */
2001 	if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2002 		afs_stat_v(dvnode, n_relpg);
2003 	return 1;
2004 }
2005 
2006 /*
2007  * invalidate part or all of a page
2008  * - release a page and clean up its private data if offset is 0 (indicating
2009  *   the entire page)
2010  */
2011 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
2012 				   unsigned int length)
2013 {
2014 	struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
2015 
2016 	_enter("{%lu},%u,%u", page->index, offset, length);
2017 
2018 	BUG_ON(!PageLocked(page));
2019 
2020 	/* The directory will need reloading. */
2021 	if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2022 		afs_stat_v(dvnode, n_inval);
2023 
2024 	/* we clean up only if the entire page is being invalidated */
2025 	if (offset == 0 && length == PAGE_SIZE) {
2026 		set_page_private(page, 0);
2027 		ClearPagePrivate(page);
2028 	}
2029 }
2030