xref: /openbmc/linux/fs/afs/dir.c (revision e1e38ea1)
1 /* dir.c: AFS filesystem directory handling
2  *
3  * Copyright (C) 2002, 2018 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/kernel.h>
13 #include <linux/fs.h>
14 #include <linux/namei.h>
15 #include <linux/pagemap.h>
16 #include <linux/swap.h>
17 #include <linux/ctype.h>
18 #include <linux/sched.h>
19 #include <linux/task_io_accounting_ops.h>
20 #include "internal.h"
21 #include "xdr_fs.h"
22 
23 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
24 				 unsigned int flags);
25 static int afs_dir_open(struct inode *inode, struct file *file);
26 static int afs_readdir(struct file *file, struct dir_context *ctx);
27 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags);
28 static int afs_d_delete(const struct dentry *dentry);
29 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name, int nlen,
30 				  loff_t fpos, u64 ino, unsigned dtype);
31 static int afs_lookup_filldir(struct dir_context *ctx, const char *name, int nlen,
32 			      loff_t fpos, u64 ino, unsigned dtype);
33 static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
34 		      bool excl);
35 static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
36 static int afs_rmdir(struct inode *dir, struct dentry *dentry);
37 static int afs_unlink(struct inode *dir, struct dentry *dentry);
38 static int afs_link(struct dentry *from, struct inode *dir,
39 		    struct dentry *dentry);
40 static int afs_symlink(struct inode *dir, struct dentry *dentry,
41 		       const char *content);
42 static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
43 		      struct inode *new_dir, struct dentry *new_dentry,
44 		      unsigned int flags);
45 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags);
46 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
47 				   unsigned int length);
48 
49 static int afs_dir_set_page_dirty(struct page *page)
50 {
51 	BUG(); /* This should never happen. */
52 }
53 
54 const struct file_operations afs_dir_file_operations = {
55 	.open		= afs_dir_open,
56 	.release	= afs_release,
57 	.iterate_shared	= afs_readdir,
58 	.lock		= afs_lock,
59 	.llseek		= generic_file_llseek,
60 };
61 
62 const struct inode_operations afs_dir_inode_operations = {
63 	.create		= afs_create,
64 	.lookup		= afs_lookup,
65 	.link		= afs_link,
66 	.unlink		= afs_unlink,
67 	.symlink	= afs_symlink,
68 	.mkdir		= afs_mkdir,
69 	.rmdir		= afs_rmdir,
70 	.rename		= afs_rename,
71 	.permission	= afs_permission,
72 	.getattr	= afs_getattr,
73 	.setattr	= afs_setattr,
74 	.listxattr	= afs_listxattr,
75 };
76 
77 const struct address_space_operations afs_dir_aops = {
78 	.set_page_dirty	= afs_dir_set_page_dirty,
79 	.releasepage	= afs_dir_releasepage,
80 	.invalidatepage	= afs_dir_invalidatepage,
81 };
82 
83 const struct dentry_operations afs_fs_dentry_operations = {
84 	.d_revalidate	= afs_d_revalidate,
85 	.d_delete	= afs_d_delete,
86 	.d_release	= afs_d_release,
87 	.d_automount	= afs_d_automount,
88 };
89 
90 struct afs_lookup_one_cookie {
91 	struct dir_context	ctx;
92 	struct qstr		name;
93 	bool			found;
94 	struct afs_fid		fid;
95 };
96 
97 struct afs_lookup_cookie {
98 	struct dir_context	ctx;
99 	struct qstr		name;
100 	bool			found;
101 	bool			one_only;
102 	unsigned short		nr_fids;
103 	struct afs_file_status	*statuses;
104 	struct afs_callback	*callbacks;
105 	struct afs_fid		fids[50];
106 };
107 
108 /*
109  * check that a directory page is valid
110  */
111 static bool afs_dir_check_page(struct afs_vnode *dvnode, struct page *page,
112 			       loff_t i_size)
113 {
114 	struct afs_xdr_dir_page *dbuf;
115 	loff_t latter, off;
116 	int tmp, qty;
117 
118 	/* Determine how many magic numbers there should be in this page, but
119 	 * we must take care because the directory may change size under us.
120 	 */
121 	off = page_offset(page);
122 	if (i_size <= off)
123 		goto checked;
124 
125 	latter = i_size - off;
126 	if (latter >= PAGE_SIZE)
127 		qty = PAGE_SIZE;
128 	else
129 		qty = latter;
130 	qty /= sizeof(union afs_xdr_dir_block);
131 
132 	/* check them */
133 	dbuf = kmap(page);
134 	for (tmp = 0; tmp < qty; tmp++) {
135 		if (dbuf->blocks[tmp].hdr.magic != AFS_DIR_MAGIC) {
136 			printk("kAFS: %s(%lx): bad magic %d/%d is %04hx\n",
137 			       __func__, dvnode->vfs_inode.i_ino, tmp, qty,
138 			       ntohs(dbuf->blocks[tmp].hdr.magic));
139 			trace_afs_dir_check_failed(dvnode, off, i_size);
140 			kunmap(page);
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  * open an AFS directory file
163  */
164 static int afs_dir_open(struct inode *inode, struct file *file)
165 {
166 	_enter("{%lu}", inode->i_ino);
167 
168 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
169 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
170 
171 	if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(inode)->flags))
172 		return -ENOENT;
173 
174 	return afs_open(inode, file);
175 }
176 
177 /*
178  * Read the directory into the pagecache in one go, scrubbing the previous
179  * contents.  The list of pages is returned, pinning them so that they don't
180  * get reclaimed during the iteration.
181  */
182 static struct afs_read *afs_read_dir(struct afs_vnode *dvnode, struct key *key)
183 	__acquires(&dvnode->validate_lock)
184 {
185 	struct afs_read *req;
186 	loff_t i_size;
187 	int nr_pages, nr_inline, i, n;
188 	int ret = -ENOMEM;
189 
190 retry:
191 	i_size = i_size_read(&dvnode->vfs_inode);
192 	if (i_size < 2048)
193 		return ERR_PTR(-EIO);
194 	if (i_size > 2048 * 1024)
195 		return ERR_PTR(-EFBIG);
196 
197 	_enter("%llu", i_size);
198 
199 	/* Get a request record to hold the page list.  We want to hold it
200 	 * inline if we can, but we don't want to make an order 1 allocation.
201 	 */
202 	nr_pages = (i_size + PAGE_SIZE - 1) / PAGE_SIZE;
203 	nr_inline = nr_pages;
204 	if (nr_inline > (PAGE_SIZE - sizeof(*req)) / sizeof(struct page *))
205 		nr_inline = 0;
206 
207 	req = kzalloc(sizeof(*req) + sizeof(struct page *) * nr_inline,
208 		      GFP_KERNEL);
209 	if (!req)
210 		return ERR_PTR(-ENOMEM);
211 
212 	refcount_set(&req->usage, 1);
213 	req->nr_pages = nr_pages;
214 	req->actual_len = i_size; /* May change */
215 	req->len = nr_pages * PAGE_SIZE; /* We can ask for more than there is */
216 	req->data_version = dvnode->status.data_version; /* May change */
217 	if (nr_inline > 0) {
218 		req->pages = req->array;
219 	} else {
220 		req->pages = kcalloc(nr_pages, sizeof(struct page *),
221 				     GFP_KERNEL);
222 		if (!req->pages)
223 			goto error;
224 	}
225 
226 	/* Get a list of all the pages that hold or will hold the directory
227 	 * content.  We need to fill in any gaps that we might find where the
228 	 * memory reclaimer has been at work.  If there are any gaps, we will
229 	 * need to reread the entire directory contents.
230 	 */
231 	i = 0;
232 	do {
233 		n = find_get_pages_contig(dvnode->vfs_inode.i_mapping, i,
234 					  req->nr_pages - i,
235 					  req->pages + i);
236 		_debug("find %u at %u/%u", n, i, req->nr_pages);
237 		if (n == 0) {
238 			gfp_t gfp = dvnode->vfs_inode.i_mapping->gfp_mask;
239 
240 			if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
241 				afs_stat_v(dvnode, n_inval);
242 
243 			ret = -ENOMEM;
244 			req->pages[i] = __page_cache_alloc(gfp);
245 			if (!req->pages[i])
246 				goto error;
247 			ret = add_to_page_cache_lru(req->pages[i],
248 						    dvnode->vfs_inode.i_mapping,
249 						    i, gfp);
250 			if (ret < 0)
251 				goto error;
252 
253 			set_page_private(req->pages[i], 1);
254 			SetPagePrivate(req->pages[i]);
255 			unlock_page(req->pages[i]);
256 			i++;
257 		} else {
258 			i += n;
259 		}
260 	} while (i < req->nr_pages);
261 
262 	/* If we're going to reload, we need to lock all the pages to prevent
263 	 * races.
264 	 */
265 	ret = -ERESTARTSYS;
266 	if (down_read_killable(&dvnode->validate_lock) < 0)
267 		goto error;
268 
269 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
270 		goto success;
271 
272 	up_read(&dvnode->validate_lock);
273 	if (down_write_killable(&dvnode->validate_lock) < 0)
274 		goto error;
275 
276 	if (!test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
277 		ret = afs_fetch_data(dvnode, key, req);
278 		if (ret < 0)
279 			goto error_unlock;
280 
281 		task_io_account_read(PAGE_SIZE * req->nr_pages);
282 
283 		if (req->len < req->file_size)
284 			goto content_has_grown;
285 
286 		/* Validate the data we just read. */
287 		ret = -EIO;
288 		for (i = 0; i < req->nr_pages; i++)
289 			if (!afs_dir_check_page(dvnode, req->pages[i],
290 						req->actual_len))
291 				goto error_unlock;
292 
293 		// TODO: Trim excess pages
294 
295 		set_bit(AFS_VNODE_DIR_VALID, &dvnode->flags);
296 	}
297 
298 	downgrade_write(&dvnode->validate_lock);
299 success:
300 	return req;
301 
302 error_unlock:
303 	up_write(&dvnode->validate_lock);
304 error:
305 	afs_put_read(req);
306 	_leave(" = %d", ret);
307 	return ERR_PTR(ret);
308 
309 content_has_grown:
310 	up_write(&dvnode->validate_lock);
311 	afs_put_read(req);
312 	goto retry;
313 }
314 
315 /*
316  * deal with one block in an AFS directory
317  */
318 static int afs_dir_iterate_block(struct dir_context *ctx,
319 				 union afs_xdr_dir_block *block,
320 				 unsigned blkoff)
321 {
322 	union afs_xdr_dirent *dire;
323 	unsigned offset, next, curr;
324 	size_t nlen;
325 	int tmp;
326 
327 	_enter("%u,%x,%p,,",(unsigned)ctx->pos,blkoff,block);
328 
329 	curr = (ctx->pos - blkoff) / sizeof(union afs_xdr_dirent);
330 
331 	/* walk through the block, an entry at a time */
332 	for (offset = (blkoff == 0 ? AFS_DIR_RESV_BLOCKS0 : AFS_DIR_RESV_BLOCKS);
333 	     offset < AFS_DIR_SLOTS_PER_BLOCK;
334 	     offset = next
335 	     ) {
336 		next = offset + 1;
337 
338 		/* skip entries marked unused in the bitmap */
339 		if (!(block->hdr.bitmap[offset / 8] &
340 		      (1 << (offset % 8)))) {
341 			_debug("ENT[%zu.%u]: unused",
342 			       blkoff / sizeof(union afs_xdr_dir_block), offset);
343 			if (offset >= curr)
344 				ctx->pos = blkoff +
345 					next * sizeof(union afs_xdr_dirent);
346 			continue;
347 		}
348 
349 		/* got a valid entry */
350 		dire = &block->dirents[offset];
351 		nlen = strnlen(dire->u.name,
352 			       sizeof(*block) -
353 			       offset * sizeof(union afs_xdr_dirent));
354 
355 		_debug("ENT[%zu.%u]: %s %zu \"%s\"",
356 		       blkoff / sizeof(union afs_xdr_dir_block), offset,
357 		       (offset < curr ? "skip" : "fill"),
358 		       nlen, dire->u.name);
359 
360 		/* work out where the next possible entry is */
361 		for (tmp = nlen; tmp > 15; tmp -= sizeof(union afs_xdr_dirent)) {
362 			if (next >= AFS_DIR_SLOTS_PER_BLOCK) {
363 				_debug("ENT[%zu.%u]:"
364 				       " %u travelled beyond end dir block"
365 				       " (len %u/%zu)",
366 				       blkoff / sizeof(union afs_xdr_dir_block),
367 				       offset, next, tmp, nlen);
368 				return -EIO;
369 			}
370 			if (!(block->hdr.bitmap[next / 8] &
371 			      (1 << (next % 8)))) {
372 				_debug("ENT[%zu.%u]:"
373 				       " %u unmarked extension (len %u/%zu)",
374 				       blkoff / sizeof(union afs_xdr_dir_block),
375 				       offset, next, tmp, nlen);
376 				return -EIO;
377 			}
378 
379 			_debug("ENT[%zu.%u]: ext %u/%zu",
380 			       blkoff / sizeof(union afs_xdr_dir_block),
381 			       next, tmp, nlen);
382 			next++;
383 		}
384 
385 		/* skip if starts before the current position */
386 		if (offset < curr)
387 			continue;
388 
389 		/* found the next entry */
390 		if (!dir_emit(ctx, dire->u.name, nlen,
391 			      ntohl(dire->u.vnode),
392 			      (ctx->actor == afs_lookup_filldir ||
393 			       ctx->actor == afs_lookup_one_filldir)?
394 			      ntohl(dire->u.unique) : DT_UNKNOWN)) {
395 			_leave(" = 0 [full]");
396 			return 0;
397 		}
398 
399 		ctx->pos = blkoff + next * sizeof(union afs_xdr_dirent);
400 	}
401 
402 	_leave(" = 1 [more]");
403 	return 1;
404 }
405 
406 /*
407  * iterate through the data blob that lists the contents of an AFS directory
408  */
409 static int afs_dir_iterate(struct inode *dir, struct dir_context *ctx,
410 			   struct key *key)
411 {
412 	struct afs_vnode *dvnode = AFS_FS_I(dir);
413 	struct afs_xdr_dir_page *dbuf;
414 	union afs_xdr_dir_block *dblock;
415 	struct afs_read *req;
416 	struct page *page;
417 	unsigned blkoff, limit;
418 	int ret;
419 
420 	_enter("{%lu},%u,,", dir->i_ino, (unsigned)ctx->pos);
421 
422 	if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(dir)->flags)) {
423 		_leave(" = -ESTALE");
424 		return -ESTALE;
425 	}
426 
427 	req = afs_read_dir(dvnode, key);
428 	if (IS_ERR(req))
429 		return PTR_ERR(req);
430 
431 	/* round the file position up to the next entry boundary */
432 	ctx->pos += sizeof(union afs_xdr_dirent) - 1;
433 	ctx->pos &= ~(sizeof(union afs_xdr_dirent) - 1);
434 
435 	/* walk through the blocks in sequence */
436 	ret = 0;
437 	while (ctx->pos < req->actual_len) {
438 		blkoff = ctx->pos & ~(sizeof(union afs_xdr_dir_block) - 1);
439 
440 		/* Fetch the appropriate page from the directory and re-add it
441 		 * to the LRU.
442 		 */
443 		page = req->pages[blkoff / PAGE_SIZE];
444 		if (!page) {
445 			ret = -EIO;
446 			break;
447 		}
448 		mark_page_accessed(page);
449 
450 		limit = blkoff & ~(PAGE_SIZE - 1);
451 
452 		dbuf = kmap(page);
453 
454 		/* deal with the individual blocks stashed on this page */
455 		do {
456 			dblock = &dbuf->blocks[(blkoff % PAGE_SIZE) /
457 					       sizeof(union afs_xdr_dir_block)];
458 			ret = afs_dir_iterate_block(ctx, dblock, blkoff);
459 			if (ret != 1) {
460 				kunmap(page);
461 				goto out;
462 			}
463 
464 			blkoff += sizeof(union afs_xdr_dir_block);
465 
466 		} while (ctx->pos < dir->i_size && blkoff < limit);
467 
468 		kunmap(page);
469 		ret = 0;
470 	}
471 
472 out:
473 	up_read(&dvnode->validate_lock);
474 	afs_put_read(req);
475 	_leave(" = %d", ret);
476 	return ret;
477 }
478 
479 /*
480  * read an AFS directory
481  */
482 static int afs_readdir(struct file *file, struct dir_context *ctx)
483 {
484 	return afs_dir_iterate(file_inode(file), ctx, afs_file_key(file));
485 }
486 
487 /*
488  * Search the directory for a single name
489  * - if afs_dir_iterate_block() spots this function, it'll pass the FID
490  *   uniquifier through dtype
491  */
492 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name,
493 				  int nlen, loff_t fpos, u64 ino, unsigned dtype)
494 {
495 	struct afs_lookup_one_cookie *cookie =
496 		container_of(ctx, struct afs_lookup_one_cookie, ctx);
497 
498 	_enter("{%s,%u},%s,%u,,%llu,%u",
499 	       cookie->name.name, cookie->name.len, name, nlen,
500 	       (unsigned long long) ino, dtype);
501 
502 	/* insanity checks first */
503 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
504 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
505 
506 	if (cookie->name.len != nlen ||
507 	    memcmp(cookie->name.name, name, nlen) != 0) {
508 		_leave(" = 0 [no]");
509 		return 0;
510 	}
511 
512 	cookie->fid.vnode = ino;
513 	cookie->fid.unique = dtype;
514 	cookie->found = 1;
515 
516 	_leave(" = -1 [found]");
517 	return -1;
518 }
519 
520 /*
521  * Do a lookup of a single name in a directory
522  * - just returns the FID the dentry name maps to if found
523  */
524 static int afs_do_lookup_one(struct inode *dir, struct dentry *dentry,
525 			     struct afs_fid *fid, struct key *key)
526 {
527 	struct afs_super_info *as = dir->i_sb->s_fs_info;
528 	struct afs_lookup_one_cookie cookie = {
529 		.ctx.actor = afs_lookup_one_filldir,
530 		.name = dentry->d_name,
531 		.fid.vid = as->volume->vid
532 	};
533 	int ret;
534 
535 	_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
536 
537 	/* search the directory */
538 	ret = afs_dir_iterate(dir, &cookie.ctx, key);
539 	if (ret < 0) {
540 		_leave(" = %d [iter]", ret);
541 		return ret;
542 	}
543 
544 	ret = -ENOENT;
545 	if (!cookie.found) {
546 		_leave(" = -ENOENT [not found]");
547 		return -ENOENT;
548 	}
549 
550 	*fid = cookie.fid;
551 	_leave(" = 0 { vn=%u u=%u }", fid->vnode, fid->unique);
552 	return 0;
553 }
554 
555 /*
556  * search the directory for a name
557  * - if afs_dir_iterate_block() spots this function, it'll pass the FID
558  *   uniquifier through dtype
559  */
560 static int afs_lookup_filldir(struct dir_context *ctx, const char *name,
561 			      int nlen, loff_t fpos, u64 ino, unsigned dtype)
562 {
563 	struct afs_lookup_cookie *cookie =
564 		container_of(ctx, struct afs_lookup_cookie, ctx);
565 	int ret;
566 
567 	_enter("{%s,%u},%s,%u,,%llu,%u",
568 	       cookie->name.name, cookie->name.len, name, nlen,
569 	       (unsigned long long) ino, dtype);
570 
571 	/* insanity checks first */
572 	BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
573 	BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
574 
575 	if (cookie->found) {
576 		if (cookie->nr_fids < 50) {
577 			cookie->fids[cookie->nr_fids].vnode	= ino;
578 			cookie->fids[cookie->nr_fids].unique	= dtype;
579 			cookie->nr_fids++;
580 		}
581 	} else if (cookie->name.len == nlen &&
582 		   memcmp(cookie->name.name, name, nlen) == 0) {
583 		cookie->fids[0].vnode	= ino;
584 		cookie->fids[0].unique	= dtype;
585 		cookie->found = 1;
586 		if (cookie->one_only)
587 			return -1;
588 	}
589 
590 	ret = cookie->nr_fids >= 50 ? -1 : 0;
591 	_leave(" = %d", ret);
592 	return ret;
593 }
594 
595 /*
596  * Do a lookup in a directory.  We make use of bulk lookup to query a slew of
597  * files in one go and create inodes for them.  The inode of the file we were
598  * asked for is returned.
599  */
600 static struct inode *afs_do_lookup(struct inode *dir, struct dentry *dentry,
601 				   struct key *key)
602 {
603 	struct afs_lookup_cookie *cookie;
604 	struct afs_cb_interest *cbi = NULL;
605 	struct afs_super_info *as = dir->i_sb->s_fs_info;
606 	struct afs_iget_data data;
607 	struct afs_fs_cursor fc;
608 	struct afs_vnode *dvnode = AFS_FS_I(dir);
609 	struct inode *inode = NULL;
610 	int ret, i;
611 
612 	_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
613 
614 	cookie = kzalloc(sizeof(struct afs_lookup_cookie), GFP_KERNEL);
615 	if (!cookie)
616 		return ERR_PTR(-ENOMEM);
617 
618 	cookie->ctx.actor = afs_lookup_filldir;
619 	cookie->name = dentry->d_name;
620 	cookie->nr_fids = 1; /* slot 0 is saved for the fid we actually want */
621 
622 	read_seqlock_excl(&dvnode->cb_lock);
623 	if (dvnode->cb_interest &&
624 	    dvnode->cb_interest->server &&
625 	    test_bit(AFS_SERVER_FL_NO_IBULK, &dvnode->cb_interest->server->flags))
626 		cookie->one_only = true;
627 	read_sequnlock_excl(&dvnode->cb_lock);
628 
629 	for (i = 0; i < 50; i++)
630 		cookie->fids[i].vid = as->volume->vid;
631 
632 	/* search the directory */
633 	ret = afs_dir_iterate(dir, &cookie->ctx, key);
634 	if (ret < 0) {
635 		inode = ERR_PTR(ret);
636 		goto out;
637 	}
638 
639 	inode = ERR_PTR(-ENOENT);
640 	if (!cookie->found)
641 		goto out;
642 
643 	/* Check to see if we already have an inode for the primary fid. */
644 	data.volume = dvnode->volume;
645 	data.fid = cookie->fids[0];
646 	inode = ilookup5(dir->i_sb, cookie->fids[0].vnode, afs_iget5_test, &data);
647 	if (inode)
648 		goto out;
649 
650 	/* Need space for examining all the selected files */
651 	inode = ERR_PTR(-ENOMEM);
652 	cookie->statuses = kcalloc(cookie->nr_fids, sizeof(struct afs_file_status),
653 				   GFP_KERNEL);
654 	if (!cookie->statuses)
655 		goto out;
656 
657 	cookie->callbacks = kcalloc(cookie->nr_fids, sizeof(struct afs_callback),
658 				    GFP_KERNEL);
659 	if (!cookie->callbacks)
660 		goto out_s;
661 
662 	/* Try FS.InlineBulkStatus first.  Abort codes for the individual
663 	 * lookups contained therein are stored in the reply without aborting
664 	 * the whole operation.
665 	 */
666 	if (cookie->one_only)
667 		goto no_inline_bulk_status;
668 
669 	inode = ERR_PTR(-ERESTARTSYS);
670 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
671 		while (afs_select_fileserver(&fc)) {
672 			if (test_bit(AFS_SERVER_FL_NO_IBULK,
673 				      &fc.cbi->server->flags)) {
674 				fc.ac.abort_code = RX_INVALID_OPERATION;
675 				fc.ac.error = -ECONNABORTED;
676 				break;
677 			}
678 			afs_fs_inline_bulk_status(&fc,
679 						  afs_v2net(dvnode),
680 						  cookie->fids,
681 						  cookie->statuses,
682 						  cookie->callbacks,
683 						  cookie->nr_fids, NULL);
684 		}
685 
686 		if (fc.ac.error == 0)
687 			cbi = afs_get_cb_interest(fc.cbi);
688 		if (fc.ac.abort_code == RX_INVALID_OPERATION)
689 			set_bit(AFS_SERVER_FL_NO_IBULK, &fc.cbi->server->flags);
690 		inode = ERR_PTR(afs_end_vnode_operation(&fc));
691 	}
692 
693 	if (!IS_ERR(inode))
694 		goto success;
695 	if (fc.ac.abort_code != RX_INVALID_OPERATION)
696 		goto out_c;
697 
698 no_inline_bulk_status:
699 	/* We could try FS.BulkStatus next, but this aborts the entire op if
700 	 * any of the lookups fails - so, for the moment, revert to
701 	 * FS.FetchStatus for just the primary fid.
702 	 */
703 	cookie->nr_fids = 1;
704 	inode = ERR_PTR(-ERESTARTSYS);
705 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
706 		while (afs_select_fileserver(&fc)) {
707 			afs_fs_fetch_status(&fc,
708 					    afs_v2net(dvnode),
709 					    cookie->fids,
710 					    cookie->statuses,
711 					    cookie->callbacks,
712 					    NULL);
713 		}
714 
715 		if (fc.ac.error == 0)
716 			cbi = afs_get_cb_interest(fc.cbi);
717 		inode = ERR_PTR(afs_end_vnode_operation(&fc));
718 	}
719 
720 	if (IS_ERR(inode))
721 		goto out_c;
722 
723 	for (i = 0; i < cookie->nr_fids; i++)
724 		cookie->statuses[i].abort_code = 0;
725 
726 success:
727 	/* Turn all the files into inodes and save the first one - which is the
728 	 * one we actually want.
729 	 */
730 	if (cookie->statuses[0].abort_code != 0)
731 		inode = ERR_PTR(afs_abort_to_error(cookie->statuses[0].abort_code));
732 
733 	for (i = 0; i < cookie->nr_fids; i++) {
734 		struct inode *ti;
735 
736 		if (cookie->statuses[i].abort_code != 0)
737 			continue;
738 
739 		ti = afs_iget(dir->i_sb, key, &cookie->fids[i],
740 			      &cookie->statuses[i],
741 			      &cookie->callbacks[i],
742 			      cbi);
743 		if (i == 0) {
744 			inode = ti;
745 		} else {
746 			if (!IS_ERR(ti))
747 				iput(ti);
748 		}
749 	}
750 
751 out_c:
752 	afs_put_cb_interest(afs_v2net(dvnode), cbi);
753 	kfree(cookie->callbacks);
754 out_s:
755 	kfree(cookie->statuses);
756 out:
757 	kfree(cookie);
758 	return inode;
759 }
760 
761 /*
762  * Look up an entry in a directory with @sys substitution.
763  */
764 static struct dentry *afs_lookup_atsys(struct inode *dir, struct dentry *dentry,
765 				       struct key *key)
766 {
767 	struct afs_sysnames *subs;
768 	struct afs_net *net = afs_i2net(dir);
769 	struct dentry *ret;
770 	char *buf, *p, *name;
771 	int len, i;
772 
773 	_enter("");
774 
775 	ret = ERR_PTR(-ENOMEM);
776 	p = buf = kmalloc(AFSNAMEMAX, GFP_KERNEL);
777 	if (!buf)
778 		goto out_p;
779 	if (dentry->d_name.len > 4) {
780 		memcpy(p, dentry->d_name.name, dentry->d_name.len - 4);
781 		p += dentry->d_name.len - 4;
782 	}
783 
784 	/* There is an ordered list of substitutes that we have to try. */
785 	read_lock(&net->sysnames_lock);
786 	subs = net->sysnames;
787 	refcount_inc(&subs->usage);
788 	read_unlock(&net->sysnames_lock);
789 
790 	for (i = 0; i < subs->nr; i++) {
791 		name = subs->subs[i];
792 		len = dentry->d_name.len - 4 + strlen(name);
793 		if (len >= AFSNAMEMAX) {
794 			ret = ERR_PTR(-ENAMETOOLONG);
795 			goto out_s;
796 		}
797 
798 		strcpy(p, name);
799 		ret = lookup_one_len(buf, dentry->d_parent, len);
800 		if (IS_ERR(ret) || d_is_positive(ret))
801 			goto out_s;
802 		dput(ret);
803 	}
804 
805 	/* We don't want to d_add() the @sys dentry here as we don't want to
806 	 * the cached dentry to hide changes to the sysnames list.
807 	 */
808 	ret = NULL;
809 out_s:
810 	afs_put_sysnames(subs);
811 	kfree(buf);
812 out_p:
813 	key_put(key);
814 	return ret;
815 }
816 
817 /*
818  * look up an entry in a directory
819  */
820 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
821 				 unsigned int flags)
822 {
823 	struct afs_vnode *dvnode = AFS_FS_I(dir);
824 	struct inode *inode;
825 	struct dentry *d;
826 	struct key *key;
827 	int ret;
828 
829 	_enter("{%x:%u},%p{%pd},",
830 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, dentry);
831 
832 	ASSERTCMP(d_inode(dentry), ==, NULL);
833 
834 	if (dentry->d_name.len >= AFSNAMEMAX) {
835 		_leave(" = -ENAMETOOLONG");
836 		return ERR_PTR(-ENAMETOOLONG);
837 	}
838 
839 	if (test_bit(AFS_VNODE_DELETED, &dvnode->flags)) {
840 		_leave(" = -ESTALE");
841 		return ERR_PTR(-ESTALE);
842 	}
843 
844 	key = afs_request_key(dvnode->volume->cell);
845 	if (IS_ERR(key)) {
846 		_leave(" = %ld [key]", PTR_ERR(key));
847 		return ERR_CAST(key);
848 	}
849 
850 	ret = afs_validate(dvnode, key);
851 	if (ret < 0) {
852 		key_put(key);
853 		_leave(" = %d [val]", ret);
854 		return ERR_PTR(ret);
855 	}
856 
857 	if (dentry->d_name.len >= 4 &&
858 	    dentry->d_name.name[dentry->d_name.len - 4] == '@' &&
859 	    dentry->d_name.name[dentry->d_name.len - 3] == 's' &&
860 	    dentry->d_name.name[dentry->d_name.len - 2] == 'y' &&
861 	    dentry->d_name.name[dentry->d_name.len - 1] == 's')
862 		return afs_lookup_atsys(dir, dentry, key);
863 
864 	afs_stat_v(dvnode, n_lookup);
865 	inode = afs_do_lookup(dir, dentry, key);
866 	key_put(key);
867 	if (inode == ERR_PTR(-ENOENT)) {
868 		inode = afs_try_auto_mntpt(dentry, dir);
869 	} else {
870 		dentry->d_fsdata =
871 			(void *)(unsigned long)dvnode->status.data_version;
872 	}
873 	d = d_splice_alias(inode, dentry);
874 	if (!IS_ERR_OR_NULL(d))
875 		d->d_fsdata = dentry->d_fsdata;
876 	return d;
877 }
878 
879 /*
880  * check that a dentry lookup hit has found a valid entry
881  * - NOTE! the hit can be a negative hit too, so we can't assume we have an
882  *   inode
883  */
884 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags)
885 {
886 	struct afs_vnode *vnode, *dir;
887 	struct afs_fid uninitialized_var(fid);
888 	struct dentry *parent;
889 	struct inode *inode;
890 	struct key *key;
891 	long dir_version, de_version;
892 	int ret;
893 
894 	if (flags & LOOKUP_RCU)
895 		return -ECHILD;
896 
897 	if (d_really_is_positive(dentry)) {
898 		vnode = AFS_FS_I(d_inode(dentry));
899 		_enter("{v={%x:%u} n=%pd fl=%lx},",
900 		       vnode->fid.vid, vnode->fid.vnode, dentry,
901 		       vnode->flags);
902 	} else {
903 		_enter("{neg n=%pd}", dentry);
904 	}
905 
906 	key = afs_request_key(AFS_FS_S(dentry->d_sb)->volume->cell);
907 	if (IS_ERR(key))
908 		key = NULL;
909 
910 	if (d_really_is_positive(dentry)) {
911 		inode = d_inode(dentry);
912 		if (inode) {
913 			vnode = AFS_FS_I(inode);
914 			afs_validate(vnode, key);
915 			if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
916 				goto out_bad;
917 		}
918 	}
919 
920 	/* lock down the parent dentry so we can peer at it */
921 	parent = dget_parent(dentry);
922 	dir = AFS_FS_I(d_inode(parent));
923 
924 	/* validate the parent directory */
925 	afs_validate(dir, key);
926 
927 	if (test_bit(AFS_VNODE_DELETED, &dir->flags)) {
928 		_debug("%pd: parent dir deleted", dentry);
929 		goto out_bad_parent;
930 	}
931 
932 	/* We only need to invalidate a dentry if the server's copy changed
933 	 * behind our back.  If we made the change, it's no problem.  Note that
934 	 * on a 32-bit system, we only have 32 bits in the dentry to store the
935 	 * version.
936 	 */
937 	dir_version = (long)dir->status.data_version;
938 	de_version = (long)dentry->d_fsdata;
939 	if (de_version == dir_version)
940 		goto out_valid;
941 
942 	dir_version = (long)dir->invalid_before;
943 	if (de_version - dir_version >= 0)
944 		goto out_valid;
945 
946 	_debug("dir modified");
947 	afs_stat_v(dir, n_reval);
948 
949 	/* search the directory for this vnode */
950 	ret = afs_do_lookup_one(&dir->vfs_inode, dentry, &fid, key);
951 	switch (ret) {
952 	case 0:
953 		/* the filename maps to something */
954 		if (d_really_is_negative(dentry))
955 			goto out_bad_parent;
956 		inode = d_inode(dentry);
957 		if (is_bad_inode(inode)) {
958 			printk("kAFS: afs_d_revalidate: %pd2 has bad inode\n",
959 			       dentry);
960 			goto out_bad_parent;
961 		}
962 
963 		vnode = AFS_FS_I(inode);
964 
965 		/* if the vnode ID has changed, then the dirent points to a
966 		 * different file */
967 		if (fid.vnode != vnode->fid.vnode) {
968 			_debug("%pd: dirent changed [%u != %u]",
969 			       dentry, fid.vnode,
970 			       vnode->fid.vnode);
971 			goto not_found;
972 		}
973 
974 		/* if the vnode ID uniqifier has changed, then the file has
975 		 * been deleted and replaced, and the original vnode ID has
976 		 * been reused */
977 		if (fid.unique != vnode->fid.unique) {
978 			_debug("%pd: file deleted (uq %u -> %u I:%u)",
979 			       dentry, fid.unique,
980 			       vnode->fid.unique,
981 			       vnode->vfs_inode.i_generation);
982 			write_seqlock(&vnode->cb_lock);
983 			set_bit(AFS_VNODE_DELETED, &vnode->flags);
984 			write_sequnlock(&vnode->cb_lock);
985 			goto not_found;
986 		}
987 		goto out_valid;
988 
989 	case -ENOENT:
990 		/* the filename is unknown */
991 		_debug("%pd: dirent not found", dentry);
992 		if (d_really_is_positive(dentry))
993 			goto not_found;
994 		goto out_valid;
995 
996 	default:
997 		_debug("failed to iterate dir %pd: %d",
998 		       parent, ret);
999 		goto out_bad_parent;
1000 	}
1001 
1002 out_valid:
1003 	dentry->d_fsdata = (void *)dir_version;
1004 	dput(parent);
1005 	key_put(key);
1006 	_leave(" = 1 [valid]");
1007 	return 1;
1008 
1009 	/* the dirent, if it exists, now points to a different vnode */
1010 not_found:
1011 	spin_lock(&dentry->d_lock);
1012 	dentry->d_flags |= DCACHE_NFSFS_RENAMED;
1013 	spin_unlock(&dentry->d_lock);
1014 
1015 out_bad_parent:
1016 	_debug("dropping dentry %pd2", dentry);
1017 	dput(parent);
1018 out_bad:
1019 	key_put(key);
1020 
1021 	_leave(" = 0 [bad]");
1022 	return 0;
1023 }
1024 
1025 /*
1026  * allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
1027  * sleep)
1028  * - called from dput() when d_count is going to 0.
1029  * - return 1 to request dentry be unhashed, 0 otherwise
1030  */
1031 static int afs_d_delete(const struct dentry *dentry)
1032 {
1033 	_enter("%pd", dentry);
1034 
1035 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1036 		goto zap;
1037 
1038 	if (d_really_is_positive(dentry) &&
1039 	    (test_bit(AFS_VNODE_DELETED,   &AFS_FS_I(d_inode(dentry))->flags) ||
1040 	     test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(d_inode(dentry))->flags)))
1041 		goto zap;
1042 
1043 	_leave(" = 0 [keep]");
1044 	return 0;
1045 
1046 zap:
1047 	_leave(" = 1 [zap]");
1048 	return 1;
1049 }
1050 
1051 /*
1052  * handle dentry release
1053  */
1054 void afs_d_release(struct dentry *dentry)
1055 {
1056 	_enter("%pd", dentry);
1057 }
1058 
1059 /*
1060  * Create a new inode for create/mkdir/symlink
1061  */
1062 static void afs_vnode_new_inode(struct afs_fs_cursor *fc,
1063 				struct dentry *new_dentry,
1064 				struct afs_fid *newfid,
1065 				struct afs_file_status *newstatus,
1066 				struct afs_callback *newcb)
1067 {
1068 	struct afs_vnode *vnode;
1069 	struct inode *inode;
1070 
1071 	if (fc->ac.error < 0)
1072 		return;
1073 
1074 	d_drop(new_dentry);
1075 
1076 	inode = afs_iget(fc->vnode->vfs_inode.i_sb, fc->key,
1077 			 newfid, newstatus, newcb, fc->cbi);
1078 	if (IS_ERR(inode)) {
1079 		/* ENOMEM or EINTR at a really inconvenient time - just abandon
1080 		 * the new directory on the server.
1081 		 */
1082 		fc->ac.error = PTR_ERR(inode);
1083 		return;
1084 	}
1085 
1086 	vnode = AFS_FS_I(inode);
1087 	set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
1088 	d_add(new_dentry, inode);
1089 }
1090 
1091 /*
1092  * create a directory on an AFS filesystem
1093  */
1094 static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1095 {
1096 	struct afs_file_status newstatus;
1097 	struct afs_fs_cursor fc;
1098 	struct afs_callback newcb;
1099 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1100 	struct afs_fid newfid;
1101 	struct key *key;
1102 	u64 data_version = dvnode->status.data_version;
1103 	int ret;
1104 
1105 	mode |= S_IFDIR;
1106 
1107 	_enter("{%x:%u},{%pd},%ho",
1108 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1109 
1110 	key = afs_request_key(dvnode->volume->cell);
1111 	if (IS_ERR(key)) {
1112 		ret = PTR_ERR(key);
1113 		goto error;
1114 	}
1115 
1116 	ret = -ERESTARTSYS;
1117 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1118 		while (afs_select_fileserver(&fc)) {
1119 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1120 			afs_fs_create(&fc, dentry->d_name.name, mode, data_version,
1121 				      &newfid, &newstatus, &newcb);
1122 		}
1123 
1124 		afs_check_for_remote_deletion(&fc, fc.vnode);
1125 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1126 		afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, &newcb);
1127 		ret = afs_end_vnode_operation(&fc);
1128 		if (ret < 0)
1129 			goto error_key;
1130 	} else {
1131 		goto error_key;
1132 	}
1133 
1134 	if (ret == 0 &&
1135 	    test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1136 		afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
1137 				 afs_edit_dir_for_create);
1138 
1139 	key_put(key);
1140 	_leave(" = 0");
1141 	return 0;
1142 
1143 error_key:
1144 	key_put(key);
1145 error:
1146 	d_drop(dentry);
1147 	_leave(" = %d", ret);
1148 	return ret;
1149 }
1150 
1151 /*
1152  * Remove a subdir from a directory.
1153  */
1154 static void afs_dir_remove_subdir(struct dentry *dentry)
1155 {
1156 	if (d_really_is_positive(dentry)) {
1157 		struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1158 
1159 		clear_nlink(&vnode->vfs_inode);
1160 		set_bit(AFS_VNODE_DELETED, &vnode->flags);
1161 		clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1162 		clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
1163 	}
1164 }
1165 
1166 /*
1167  * remove a directory from an AFS filesystem
1168  */
1169 static int afs_rmdir(struct inode *dir, struct dentry *dentry)
1170 {
1171 	struct afs_fs_cursor fc;
1172 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1173 	struct key *key;
1174 	u64 data_version = dvnode->status.data_version;
1175 	int ret;
1176 
1177 	_enter("{%x:%u},{%pd}",
1178 	       dvnode->fid.vid, dvnode->fid.vnode, dentry);
1179 
1180 	key = afs_request_key(dvnode->volume->cell);
1181 	if (IS_ERR(key)) {
1182 		ret = PTR_ERR(key);
1183 		goto error;
1184 	}
1185 
1186 	ret = -ERESTARTSYS;
1187 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1188 		while (afs_select_fileserver(&fc)) {
1189 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1190 			afs_fs_remove(&fc, dentry->d_name.name, true,
1191 				      data_version);
1192 		}
1193 
1194 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1195 		ret = afs_end_vnode_operation(&fc);
1196 		if (ret == 0) {
1197 			afs_dir_remove_subdir(dentry);
1198 			if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1199 				afs_edit_dir_remove(dvnode, &dentry->d_name,
1200 						    afs_edit_dir_for_rmdir);
1201 		}
1202 	}
1203 
1204 	key_put(key);
1205 error:
1206 	return ret;
1207 }
1208 
1209 /*
1210  * Remove a link to a file or symlink from a directory.
1211  *
1212  * If the file was not deleted due to excess hard links, the fileserver will
1213  * break the callback promise on the file - if it had one - before it returns
1214  * to us, and if it was deleted, it won't
1215  *
1216  * However, if we didn't have a callback promise outstanding, or it was
1217  * outstanding on a different server, then it won't break it either...
1218  */
1219 static int afs_dir_remove_link(struct dentry *dentry, struct key *key,
1220 			       unsigned long d_version_before,
1221 			       unsigned long d_version_after)
1222 {
1223 	bool dir_valid;
1224 	int ret = 0;
1225 
1226 	/* There were no intervening changes on the server if the version
1227 	 * number we got back was incremented by exactly 1.
1228 	 */
1229 	dir_valid = (d_version_after == d_version_before + 1);
1230 
1231 	if (d_really_is_positive(dentry)) {
1232 		struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1233 
1234 		if (dir_valid) {
1235 			drop_nlink(&vnode->vfs_inode);
1236 			if (vnode->vfs_inode.i_nlink == 0) {
1237 				set_bit(AFS_VNODE_DELETED, &vnode->flags);
1238 				clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1239 			}
1240 			ret = 0;
1241 		} else {
1242 			clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1243 
1244 			if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1245 				kdebug("AFS_VNODE_DELETED");
1246 
1247 			ret = afs_validate(vnode, key);
1248 			if (ret == -ESTALE)
1249 				ret = 0;
1250 		}
1251 		_debug("nlink %d [val %d]", vnode->vfs_inode.i_nlink, ret);
1252 	}
1253 
1254 	return ret;
1255 }
1256 
1257 /*
1258  * Remove a file or symlink from an AFS filesystem.
1259  */
1260 static int afs_unlink(struct inode *dir, struct dentry *dentry)
1261 {
1262 	struct afs_fs_cursor fc;
1263 	struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode;
1264 	struct key *key;
1265 	unsigned long d_version = (unsigned long)dentry->d_fsdata;
1266 	u64 data_version = dvnode->status.data_version;
1267 	int ret;
1268 
1269 	_enter("{%x:%u},{%pd}",
1270 	       dvnode->fid.vid, dvnode->fid.vnode, dentry);
1271 
1272 	if (dentry->d_name.len >= AFSNAMEMAX)
1273 		return -ENAMETOOLONG;
1274 
1275 	key = afs_request_key(dvnode->volume->cell);
1276 	if (IS_ERR(key)) {
1277 		ret = PTR_ERR(key);
1278 		goto error;
1279 	}
1280 
1281 	/* Try to make sure we have a callback promise on the victim. */
1282 	if (d_really_is_positive(dentry)) {
1283 		vnode = AFS_FS_I(d_inode(dentry));
1284 		ret = afs_validate(vnode, key);
1285 		if (ret < 0)
1286 			goto error_key;
1287 	}
1288 
1289 	ret = -ERESTARTSYS;
1290 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1291 		while (afs_select_fileserver(&fc)) {
1292 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1293 			afs_fs_remove(&fc, dentry->d_name.name, false,
1294 				      data_version);
1295 		}
1296 
1297 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1298 		ret = afs_end_vnode_operation(&fc);
1299 		if (ret == 0)
1300 			ret = afs_dir_remove_link(
1301 				dentry, key, d_version,
1302 				(unsigned long)dvnode->status.data_version);
1303 		if (ret == 0 &&
1304 		    test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1305 			afs_edit_dir_remove(dvnode, &dentry->d_name,
1306 					    afs_edit_dir_for_unlink);
1307 	}
1308 
1309 error_key:
1310 	key_put(key);
1311 error:
1312 	_leave(" = %d", ret);
1313 	return ret;
1314 }
1315 
1316 /*
1317  * create a regular file on an AFS filesystem
1318  */
1319 static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1320 		      bool excl)
1321 {
1322 	struct afs_fs_cursor fc;
1323 	struct afs_file_status newstatus;
1324 	struct afs_callback newcb;
1325 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1326 	struct afs_fid newfid;
1327 	struct key *key;
1328 	u64 data_version = dvnode->status.data_version;
1329 	int ret;
1330 
1331 	mode |= S_IFREG;
1332 
1333 	_enter("{%x:%u},{%pd},%ho,",
1334 	       dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1335 
1336 	ret = -ENAMETOOLONG;
1337 	if (dentry->d_name.len >= AFSNAMEMAX)
1338 		goto error;
1339 
1340 	key = afs_request_key(dvnode->volume->cell);
1341 	if (IS_ERR(key)) {
1342 		ret = PTR_ERR(key);
1343 		goto error;
1344 	}
1345 
1346 	ret = -ERESTARTSYS;
1347 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1348 		while (afs_select_fileserver(&fc)) {
1349 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1350 			afs_fs_create(&fc, dentry->d_name.name, mode, data_version,
1351 				      &newfid, &newstatus, &newcb);
1352 		}
1353 
1354 		afs_check_for_remote_deletion(&fc, fc.vnode);
1355 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1356 		afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, &newcb);
1357 		ret = afs_end_vnode_operation(&fc);
1358 		if (ret < 0)
1359 			goto error_key;
1360 	} else {
1361 		goto error_key;
1362 	}
1363 
1364 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1365 		afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
1366 				 afs_edit_dir_for_create);
1367 
1368 	key_put(key);
1369 	_leave(" = 0");
1370 	return 0;
1371 
1372 error_key:
1373 	key_put(key);
1374 error:
1375 	d_drop(dentry);
1376 	_leave(" = %d", ret);
1377 	return ret;
1378 }
1379 
1380 /*
1381  * create a hard link between files in an AFS filesystem
1382  */
1383 static int afs_link(struct dentry *from, struct inode *dir,
1384 		    struct dentry *dentry)
1385 {
1386 	struct afs_fs_cursor fc;
1387 	struct afs_vnode *dvnode, *vnode;
1388 	struct key *key;
1389 	u64 data_version;
1390 	int ret;
1391 
1392 	vnode = AFS_FS_I(d_inode(from));
1393 	dvnode = AFS_FS_I(dir);
1394 	data_version = dvnode->status.data_version;
1395 
1396 	_enter("{%x:%u},{%x:%u},{%pd}",
1397 	       vnode->fid.vid, vnode->fid.vnode,
1398 	       dvnode->fid.vid, dvnode->fid.vnode,
1399 	       dentry);
1400 
1401 	ret = -ENAMETOOLONG;
1402 	if (dentry->d_name.len >= AFSNAMEMAX)
1403 		goto error;
1404 
1405 	key = afs_request_key(dvnode->volume->cell);
1406 	if (IS_ERR(key)) {
1407 		ret = PTR_ERR(key);
1408 		goto error;
1409 	}
1410 
1411 	ret = -ERESTARTSYS;
1412 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1413 		if (mutex_lock_interruptible_nested(&vnode->io_lock, 1) < 0) {
1414 			afs_end_vnode_operation(&fc);
1415 			goto error_key;
1416 		}
1417 
1418 		while (afs_select_fileserver(&fc)) {
1419 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1420 			fc.cb_break_2 = afs_calc_vnode_cb_break(vnode);
1421 			afs_fs_link(&fc, vnode, dentry->d_name.name, data_version);
1422 		}
1423 
1424 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1425 		afs_vnode_commit_status(&fc, vnode, fc.cb_break_2);
1426 		ihold(&vnode->vfs_inode);
1427 		d_instantiate(dentry, &vnode->vfs_inode);
1428 
1429 		mutex_unlock(&vnode->io_lock);
1430 		ret = afs_end_vnode_operation(&fc);
1431 		if (ret < 0)
1432 			goto error_key;
1433 	} else {
1434 		goto error_key;
1435 	}
1436 
1437 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1438 		afs_edit_dir_add(dvnode, &dentry->d_name, &vnode->fid,
1439 				 afs_edit_dir_for_link);
1440 
1441 	key_put(key);
1442 	_leave(" = 0");
1443 	return 0;
1444 
1445 error_key:
1446 	key_put(key);
1447 error:
1448 	d_drop(dentry);
1449 	_leave(" = %d", ret);
1450 	return ret;
1451 }
1452 
1453 /*
1454  * create a symlink in an AFS filesystem
1455  */
1456 static int afs_symlink(struct inode *dir, struct dentry *dentry,
1457 		       const char *content)
1458 {
1459 	struct afs_fs_cursor fc;
1460 	struct afs_file_status newstatus;
1461 	struct afs_vnode *dvnode = AFS_FS_I(dir);
1462 	struct afs_fid newfid;
1463 	struct key *key;
1464 	u64 data_version = dvnode->status.data_version;
1465 	int ret;
1466 
1467 	_enter("{%x:%u},{%pd},%s",
1468 	       dvnode->fid.vid, dvnode->fid.vnode, dentry,
1469 	       content);
1470 
1471 	ret = -ENAMETOOLONG;
1472 	if (dentry->d_name.len >= AFSNAMEMAX)
1473 		goto error;
1474 
1475 	ret = -EINVAL;
1476 	if (strlen(content) >= AFSPATHMAX)
1477 		goto error;
1478 
1479 	key = afs_request_key(dvnode->volume->cell);
1480 	if (IS_ERR(key)) {
1481 		ret = PTR_ERR(key);
1482 		goto error;
1483 	}
1484 
1485 	ret = -ERESTARTSYS;
1486 	if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1487 		while (afs_select_fileserver(&fc)) {
1488 			fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1489 			afs_fs_symlink(&fc, dentry->d_name.name,
1490 				       content, data_version,
1491 				       &newfid, &newstatus);
1492 		}
1493 
1494 		afs_check_for_remote_deletion(&fc, fc.vnode);
1495 		afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1496 		afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, NULL);
1497 		ret = afs_end_vnode_operation(&fc);
1498 		if (ret < 0)
1499 			goto error_key;
1500 	} else {
1501 		goto error_key;
1502 	}
1503 
1504 	if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1505 		afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
1506 				 afs_edit_dir_for_symlink);
1507 
1508 	key_put(key);
1509 	_leave(" = 0");
1510 	return 0;
1511 
1512 error_key:
1513 	key_put(key);
1514 error:
1515 	d_drop(dentry);
1516 	_leave(" = %d", ret);
1517 	return ret;
1518 }
1519 
1520 /*
1521  * rename a file in an AFS filesystem and/or move it between directories
1522  */
1523 static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
1524 		      struct inode *new_dir, struct dentry *new_dentry,
1525 		      unsigned int flags)
1526 {
1527 	struct afs_fs_cursor fc;
1528 	struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
1529 	struct key *key;
1530 	u64 orig_data_version, new_data_version;
1531 	bool new_negative = d_is_negative(new_dentry);
1532 	int ret;
1533 
1534 	if (flags)
1535 		return -EINVAL;
1536 
1537 	vnode = AFS_FS_I(d_inode(old_dentry));
1538 	orig_dvnode = AFS_FS_I(old_dir);
1539 	new_dvnode = AFS_FS_I(new_dir);
1540 	orig_data_version = orig_dvnode->status.data_version;
1541 	new_data_version = new_dvnode->status.data_version;
1542 
1543 	_enter("{%x:%u},{%x:%u},{%x:%u},{%pd}",
1544 	       orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
1545 	       vnode->fid.vid, vnode->fid.vnode,
1546 	       new_dvnode->fid.vid, new_dvnode->fid.vnode,
1547 	       new_dentry);
1548 
1549 	key = afs_request_key(orig_dvnode->volume->cell);
1550 	if (IS_ERR(key)) {
1551 		ret = PTR_ERR(key);
1552 		goto error;
1553 	}
1554 
1555 	ret = -ERESTARTSYS;
1556 	if (afs_begin_vnode_operation(&fc, orig_dvnode, key)) {
1557 		if (orig_dvnode != new_dvnode) {
1558 			if (mutex_lock_interruptible_nested(&new_dvnode->io_lock, 1) < 0) {
1559 				afs_end_vnode_operation(&fc);
1560 				goto error_key;
1561 			}
1562 		}
1563 		while (afs_select_fileserver(&fc)) {
1564 			fc.cb_break = afs_calc_vnode_cb_break(orig_dvnode);
1565 			fc.cb_break_2 = afs_calc_vnode_cb_break(new_dvnode);
1566 			afs_fs_rename(&fc, old_dentry->d_name.name,
1567 				      new_dvnode, new_dentry->d_name.name,
1568 				      orig_data_version, new_data_version);
1569 		}
1570 
1571 		afs_vnode_commit_status(&fc, orig_dvnode, fc.cb_break);
1572 		afs_vnode_commit_status(&fc, new_dvnode, fc.cb_break_2);
1573 		if (orig_dvnode != new_dvnode)
1574 			mutex_unlock(&new_dvnode->io_lock);
1575 		ret = afs_end_vnode_operation(&fc);
1576 		if (ret < 0)
1577 			goto error_key;
1578 	}
1579 
1580 	if (ret == 0) {
1581 		if (test_bit(AFS_VNODE_DIR_VALID, &orig_dvnode->flags))
1582 		    afs_edit_dir_remove(orig_dvnode, &old_dentry->d_name,
1583 					afs_edit_dir_for_rename);
1584 
1585 		if (!new_negative &&
1586 		    test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
1587 			afs_edit_dir_remove(new_dvnode, &new_dentry->d_name,
1588 					    afs_edit_dir_for_rename);
1589 
1590 		if (test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
1591 			afs_edit_dir_add(new_dvnode, &new_dentry->d_name,
1592 					 &vnode->fid,  afs_edit_dir_for_rename);
1593 	}
1594 
1595 error_key:
1596 	key_put(key);
1597 error:
1598 	_leave(" = %d", ret);
1599 	return ret;
1600 }
1601 
1602 /*
1603  * Release a directory page and clean up its private state if it's not busy
1604  * - return true if the page can now be released, false if not
1605  */
1606 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags)
1607 {
1608 	struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
1609 
1610 	_enter("{{%x:%u}[%lu]}", dvnode->fid.vid, dvnode->fid.vnode, page->index);
1611 
1612 	set_page_private(page, 0);
1613 	ClearPagePrivate(page);
1614 
1615 	/* The directory will need reloading. */
1616 	if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1617 		afs_stat_v(dvnode, n_relpg);
1618 	return 1;
1619 }
1620 
1621 /*
1622  * invalidate part or all of a page
1623  * - release a page and clean up its private data if offset is 0 (indicating
1624  *   the entire page)
1625  */
1626 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
1627 				   unsigned int length)
1628 {
1629 	struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
1630 
1631 	_enter("{%lu},%u,%u", page->index, offset, length);
1632 
1633 	BUG_ON(!PageLocked(page));
1634 
1635 	/* The directory will need reloading. */
1636 	if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1637 		afs_stat_v(dvnode, n_inval);
1638 
1639 	/* we clean up only if the entire page is being invalidated */
1640 	if (offset == 0 && length == PAGE_SIZE) {
1641 		set_page_private(page, 0);
1642 		ClearPagePrivate(page);
1643 	}
1644 }
1645