Revert "tty: hvc: Fix data abort due to race in hvc_open"
[linux/fpc-iii.git] / fs / afs / dir.c
blob3c486340b22085f8e23bec6780a5d0dad16468b8
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* dir.c: AFS filesystem directory handling
4 * Copyright (C) 2002, 2018 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
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"
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);
47 static int afs_dir_set_page_dirty(struct page *page)
49 BUG(); /* This should never happen. */
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,
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,
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,
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,
89 struct afs_lookup_one_cookie {
90 struct dir_context ctx;
91 struct qstr name;
92 bool found;
93 struct afs_fid fid;
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];
108 * check that a directory page is valid
110 static bool afs_dir_check_page(struct afs_vnode *dvnode, struct page *page,
111 loff_t i_size)
113 struct afs_xdr_dir_page *dbuf;
114 loff_t latter, off;
115 int tmp, qty;
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.
120 off = page_offset(page);
121 if (i_size <= off)
122 goto checked;
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);
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;
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.
148 ((u8 *)&dbuf->blocks[tmp])[AFS_DIR_BLOCK_SIZE - 1] = 0;
151 kunmap(page);
153 checked:
154 afs_stat_v(dvnode, n_read_dir);
155 return true;
157 error:
158 return false;
162 * Check the contents of a directory that we've just read.
164 static bool afs_dir_check_pages(struct afs_vnode *dvnode, struct afs_read *req)
166 struct afs_xdr_dir_page *dbuf;
167 unsigned int i, j, qty = PAGE_SIZE / sizeof(union afs_xdr_dir_block);
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;
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);
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];
186 pr_warn("[%02x] %32phN\n", i * qty + j, block);
188 kunmap(req->pages[i]);
190 return false;
194 * open an AFS directory file
196 static int afs_dir_open(struct inode *inode, struct file *file)
198 _enter("{%lu}", inode->i_ino);
200 BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
201 BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
203 if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(inode)->flags))
204 return -ENOENT;
206 return afs_open(inode, file);
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.
214 static struct afs_read *afs_read_dir(struct afs_vnode *dvnode, struct key *key)
215 __acquires(&dvnode->validate_lock)
217 struct afs_read *req;
218 loff_t i_size;
219 int nr_pages, nr_inline, i, n;
220 int ret = -ENOMEM;
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);
231 _enter("%llu", i_size);
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.
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;
241 req = kzalloc(struct_size(req, array, nr_inline), GFP_KERNEL);
242 if (!req)
243 return ERR_PTR(-ENOMEM);
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;
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.
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;
273 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
274 afs_stat_v(dvnode, n_inval);
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;
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;
293 } while (i < req->nr_pages);
295 /* If we're going to reload, we need to lock all the pages to prevent
296 * races.
298 ret = -ERESTARTSYS;
299 if (down_read_killable(&dvnode->validate_lock) < 0)
300 goto error;
302 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
303 goto success;
305 up_read(&dvnode->validate_lock);
306 if (down_write_killable(&dvnode->validate_lock) < 0)
307 goto error;
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;
315 task_io_account_read(PAGE_SIZE * req->nr_pages);
317 if (req->len < req->file_size)
318 goto content_has_grown;
320 /* Validate the data we just read. */
321 ret = -EIO;
322 if (!afs_dir_check_pages(dvnode, req))
323 goto error_unlock;
325 // TODO: Trim excess pages
327 set_bit(AFS_VNODE_DIR_VALID, &dvnode->flags);
330 downgrade_write(&dvnode->validate_lock);
331 success:
332 return req;
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);
341 content_has_grown:
342 up_write(&dvnode->validate_lock);
343 afs_put_read(req);
344 goto retry;
348 * deal with one block in an AFS directory
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)
355 union afs_xdr_dirent *dire;
356 unsigned offset, next, curr;
357 size_t nlen;
358 int tmp;
360 _enter("%u,%x,%p,,",(unsigned)ctx->pos,blkoff,block);
362 curr = (ctx->pos - blkoff) / sizeof(union afs_xdr_dirent);
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
369 next = offset + 1;
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;
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));
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);
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);
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);
412 _debug("ENT[%zu.%u]: ext %u/%zu",
413 blkoff / sizeof(union afs_xdr_dir_block),
414 next, tmp, nlen);
415 next++;
418 /* skip if starts before the current position */
419 if (offset < curr)
420 continue;
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;
432 ctx->pos = blkoff + next * sizeof(union afs_xdr_dirent);
435 _leave(" = 1 [more]");
436 return 1;
440 * iterate through the data blob that lists the contents of an AFS directory
442 static int afs_dir_iterate(struct inode *dir, struct dir_context *ctx,
443 struct key *key, afs_dataversion_t *_dir_version)
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;
453 _enter("{%lu},%u,,", dir->i_ino, (unsigned)ctx->pos);
455 if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(dir)->flags)) {
456 _leave(" = -ESTALE");
457 return -ESTALE;
460 req = afs_read_dir(dvnode, key);
461 if (IS_ERR(req))
462 return PTR_ERR(req);
463 *_dir_version = req->data_version;
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);
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);
474 /* Fetch the appropriate page from the directory and re-add it
475 * to the LRU.
477 page = req->pages[blkoff / PAGE_SIZE];
478 if (!page) {
479 ret = afs_bad(dvnode, afs_file_error_dir_missing_page);
480 break;
482 mark_page_accessed(page);
484 limit = blkoff & ~(PAGE_SIZE - 1);
486 dbuf = kmap(page);
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;
498 blkoff += sizeof(union afs_xdr_dir_block);
500 } while (ctx->pos < dir->i_size && blkoff < limit);
502 kunmap(page);
503 ret = 0;
506 out:
507 up_read(&dvnode->validate_lock);
508 afs_put_read(req);
509 _leave(" = %d", ret);
510 return ret;
514 * read an AFS directory
516 static int afs_readdir(struct file *file, struct dir_context *ctx)
518 afs_dataversion_t dir_version;
520 return afs_dir_iterate(file_inode(file), ctx, afs_file_key(file),
521 &dir_version);
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
529 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name,
530 int nlen, loff_t fpos, u64 ino, unsigned dtype)
532 struct afs_lookup_one_cookie *cookie =
533 container_of(ctx, struct afs_lookup_one_cookie, ctx);
535 _enter("{%s,%u},%s,%u,,%llu,%u",
536 cookie->name.name, cookie->name.len, name, nlen,
537 (unsigned long long) ino, dtype);
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);
543 if (cookie->name.len != nlen ||
544 memcmp(cookie->name.name, name, nlen) != 0) {
545 _leave(" = 0 [no]");
546 return 0;
549 cookie->fid.vnode = ino;
550 cookie->fid.unique = dtype;
551 cookie->found = 1;
553 _leave(" = -1 [found]");
554 return -1;
558 * Do a lookup of a single name in a directory
559 * - just returns the FID the dentry name maps to if found
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)
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
571 int ret;
573 _enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
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;
582 ret = -ENOENT;
583 if (!cookie.found) {
584 _leave(" = -ENOENT [not found]");
585 return -ENOENT;
588 *fid = cookie.fid;
589 _leave(" = 0 { vn=%llu u=%u }", fid->vnode, fid->unique);
590 return 0;
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
598 static int afs_lookup_filldir(struct dir_context *ctx, const char *name,
599 int nlen, loff_t fpos, u64 ino, unsigned dtype)
601 struct afs_lookup_cookie *cookie =
602 container_of(ctx, struct afs_lookup_cookie, ctx);
603 int ret;
605 _enter("{%s,%u},%s,%u,,%llu,%u",
606 cookie->name.name, cookie->name.len, name, nlen,
607 (unsigned long long) ino, dtype);
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);
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++;
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;
628 ret = cookie->nr_fids >= 50 ? -1 : 0;
629 _leave(" = %d", ret);
630 return ret;
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.
638 static struct inode *afs_do_lookup(struct inode *dir, struct dentry *dentry,
639 struct key *key)
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;
653 _enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
655 cookie = kzalloc(sizeof(struct afs_lookup_cookie), GFP_KERNEL);
656 if (!cookie)
657 return ERR_PTR(-ENOMEM);
659 cookie->ctx.actor = afs_lookup_filldir;
660 cookie->name = dentry->d_name;
661 cookie->nr_fids = 2; /* slot 0 is saved for the fid we actually want
662 * and slot 1 for the directory */
664 read_seqlock_excl(&dvnode->cb_lock);
665 dcbi = rcu_dereference_protected(dvnode->cb_interest,
666 lockdep_is_held(&dvnode->cb_lock.lock));
667 if (dcbi) {
668 server = dcbi->server;
669 if (server &&
670 test_bit(AFS_SERVER_FL_NO_IBULK, &server->flags))
671 cookie->one_only = true;
673 read_sequnlock_excl(&dvnode->cb_lock);
675 for (i = 0; i < 50; i++)
676 cookie->fids[i].vid = as->volume->vid;
678 /* search the directory */
679 ret = afs_dir_iterate(dir, &cookie->ctx, key, &data_version);
680 if (ret < 0) {
681 inode = ERR_PTR(ret);
682 goto out;
685 dentry->d_fsdata = (void *)(unsigned long)data_version;
687 inode = ERR_PTR(-ENOENT);
688 if (!cookie->found)
689 goto out;
691 /* Check to see if we already have an inode for the primary fid. */
692 iget_data.fid = cookie->fids[0];
693 iget_data.volume = dvnode->volume;
694 iget_data.cb_v_break = dvnode->volume->cb_v_break;
695 iget_data.cb_s_break = 0;
696 inode = ilookup5(dir->i_sb, cookie->fids[0].vnode,
697 afs_iget5_test, &iget_data);
698 if (inode)
699 goto out;
701 /* Need space for examining all the selected files */
702 inode = ERR_PTR(-ENOMEM);
703 cookie->statuses = kvcalloc(cookie->nr_fids, sizeof(struct afs_status_cb),
704 GFP_KERNEL);
705 if (!cookie->statuses)
706 goto out;
708 cookie->inodes = kcalloc(cookie->nr_fids, sizeof(struct inode *),
709 GFP_KERNEL);
710 if (!cookie->inodes)
711 goto out_s;
713 cookie->fids[1] = dvnode->fid;
714 cookie->statuses[1].cb_break = afs_calc_vnode_cb_break(dvnode);
715 cookie->inodes[1] = igrab(&dvnode->vfs_inode);
717 for (i = 2; i < cookie->nr_fids; i++) {
718 scb = &cookie->statuses[i];
720 /* Find any inodes that already exist and get their
721 * callback counters.
723 iget_data.fid = cookie->fids[i];
724 ti = ilookup5_nowait(dir->i_sb, iget_data.fid.vnode,
725 afs_iget5_test, &iget_data);
726 if (!IS_ERR_OR_NULL(ti)) {
727 vnode = AFS_FS_I(ti);
728 scb->cb_break = afs_calc_vnode_cb_break(vnode);
729 cookie->inodes[i] = ti;
733 /* Try FS.InlineBulkStatus first. Abort codes for the individual
734 * lookups contained therein are stored in the reply without aborting
735 * the whole operation.
737 if (cookie->one_only)
738 goto no_inline_bulk_status;
740 inode = ERR_PTR(-ERESTARTSYS);
741 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
742 while (afs_select_fileserver(&fc)) {
743 if (test_bit(AFS_SERVER_FL_NO_IBULK,
744 &fc.cbi->server->flags)) {
745 fc.ac.abort_code = RX_INVALID_OPERATION;
746 fc.ac.error = -ECONNABORTED;
747 break;
749 iget_data.cb_v_break = dvnode->volume->cb_v_break;
750 iget_data.cb_s_break = fc.cbi->server->cb_s_break;
751 afs_fs_inline_bulk_status(&fc,
752 afs_v2net(dvnode),
753 cookie->fids,
754 cookie->statuses,
755 cookie->nr_fids, NULL);
758 if (fc.ac.error == 0)
759 cbi = afs_get_cb_interest(fc.cbi);
760 if (fc.ac.abort_code == RX_INVALID_OPERATION)
761 set_bit(AFS_SERVER_FL_NO_IBULK, &fc.cbi->server->flags);
762 inode = ERR_PTR(afs_end_vnode_operation(&fc));
765 if (!IS_ERR(inode))
766 goto success;
767 if (fc.ac.abort_code != RX_INVALID_OPERATION)
768 goto out_c;
770 no_inline_bulk_status:
771 /* We could try FS.BulkStatus next, but this aborts the entire op if
772 * any of the lookups fails - so, for the moment, revert to
773 * FS.FetchStatus for just the primary fid.
775 inode = ERR_PTR(-ERESTARTSYS);
776 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
777 while (afs_select_fileserver(&fc)) {
778 iget_data.cb_v_break = dvnode->volume->cb_v_break;
779 iget_data.cb_s_break = fc.cbi->server->cb_s_break;
780 scb = &cookie->statuses[0];
781 afs_fs_fetch_status(&fc,
782 afs_v2net(dvnode),
783 cookie->fids,
784 scb,
785 NULL);
788 if (fc.ac.error == 0)
789 cbi = afs_get_cb_interest(fc.cbi);
790 inode = ERR_PTR(afs_end_vnode_operation(&fc));
793 if (IS_ERR(inode))
794 goto out_c;
796 success:
797 /* Turn all the files into inodes and save the first one - which is the
798 * one we actually want.
800 scb = &cookie->statuses[0];
801 if (scb->status.abort_code != 0)
802 inode = ERR_PTR(afs_abort_to_error(scb->status.abort_code));
804 for (i = 0; i < cookie->nr_fids; i++) {
805 struct afs_status_cb *scb = &cookie->statuses[i];
807 if (!scb->have_status && !scb->have_error)
808 continue;
810 if (cookie->inodes[i]) {
811 struct afs_vnode *iv = AFS_FS_I(cookie->inodes[i]);
813 if (test_bit(AFS_VNODE_UNSET, &iv->flags))
814 continue;
816 afs_vnode_commit_status(&fc, iv,
817 scb->cb_break, NULL, scb);
818 continue;
821 if (scb->status.abort_code != 0)
822 continue;
824 iget_data.fid = cookie->fids[i];
825 ti = afs_iget(dir->i_sb, key, &iget_data, scb, cbi, dvnode);
826 if (!IS_ERR(ti))
827 afs_cache_permit(AFS_FS_I(ti), key,
828 0 /* Assume vnode->cb_break is 0 */ +
829 iget_data.cb_v_break,
830 scb);
831 if (i == 0) {
832 inode = ti;
833 } else {
834 if (!IS_ERR(ti))
835 iput(ti);
839 out_c:
840 afs_put_cb_interest(afs_v2net(dvnode), cbi);
841 if (cookie->inodes) {
842 for (i = 0; i < cookie->nr_fids; i++)
843 iput(cookie->inodes[i]);
844 kfree(cookie->inodes);
846 out_s:
847 kvfree(cookie->statuses);
848 out:
849 kfree(cookie);
850 return inode;
854 * Look up an entry in a directory with @sys substitution.
856 static struct dentry *afs_lookup_atsys(struct inode *dir, struct dentry *dentry,
857 struct key *key)
859 struct afs_sysnames *subs;
860 struct afs_net *net = afs_i2net(dir);
861 struct dentry *ret;
862 char *buf, *p, *name;
863 int len, i;
865 _enter("");
867 ret = ERR_PTR(-ENOMEM);
868 p = buf = kmalloc(AFSNAMEMAX, GFP_KERNEL);
869 if (!buf)
870 goto out_p;
871 if (dentry->d_name.len > 4) {
872 memcpy(p, dentry->d_name.name, dentry->d_name.len - 4);
873 p += dentry->d_name.len - 4;
876 /* There is an ordered list of substitutes that we have to try. */
877 read_lock(&net->sysnames_lock);
878 subs = net->sysnames;
879 refcount_inc(&subs->usage);
880 read_unlock(&net->sysnames_lock);
882 for (i = 0; i < subs->nr; i++) {
883 name = subs->subs[i];
884 len = dentry->d_name.len - 4 + strlen(name);
885 if (len >= AFSNAMEMAX) {
886 ret = ERR_PTR(-ENAMETOOLONG);
887 goto out_s;
890 strcpy(p, name);
891 ret = lookup_one_len(buf, dentry->d_parent, len);
892 if (IS_ERR(ret) || d_is_positive(ret))
893 goto out_s;
894 dput(ret);
897 /* We don't want to d_add() the @sys dentry here as we don't want to
898 * the cached dentry to hide changes to the sysnames list.
900 ret = NULL;
901 out_s:
902 afs_put_sysnames(subs);
903 kfree(buf);
904 out_p:
905 key_put(key);
906 return ret;
910 * look up an entry in a directory
912 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
913 unsigned int flags)
915 struct afs_vnode *dvnode = AFS_FS_I(dir);
916 struct afs_fid fid = {};
917 struct inode *inode;
918 struct dentry *d;
919 struct key *key;
920 int ret;
922 _enter("{%llx:%llu},%p{%pd},",
923 dvnode->fid.vid, dvnode->fid.vnode, dentry, dentry);
925 ASSERTCMP(d_inode(dentry), ==, NULL);
927 if (dentry->d_name.len >= AFSNAMEMAX) {
928 _leave(" = -ENAMETOOLONG");
929 return ERR_PTR(-ENAMETOOLONG);
932 if (test_bit(AFS_VNODE_DELETED, &dvnode->flags)) {
933 _leave(" = -ESTALE");
934 return ERR_PTR(-ESTALE);
937 key = afs_request_key(dvnode->volume->cell);
938 if (IS_ERR(key)) {
939 _leave(" = %ld [key]", PTR_ERR(key));
940 return ERR_CAST(key);
943 ret = afs_validate(dvnode, key);
944 if (ret < 0) {
945 key_put(key);
946 _leave(" = %d [val]", ret);
947 return ERR_PTR(ret);
950 if (dentry->d_name.len >= 4 &&
951 dentry->d_name.name[dentry->d_name.len - 4] == '@' &&
952 dentry->d_name.name[dentry->d_name.len - 3] == 's' &&
953 dentry->d_name.name[dentry->d_name.len - 2] == 'y' &&
954 dentry->d_name.name[dentry->d_name.len - 1] == 's')
955 return afs_lookup_atsys(dir, dentry, key);
957 afs_stat_v(dvnode, n_lookup);
958 inode = afs_do_lookup(dir, dentry, key);
959 key_put(key);
960 if (inode == ERR_PTR(-ENOENT))
961 inode = afs_try_auto_mntpt(dentry, dir);
963 if (!IS_ERR_OR_NULL(inode))
964 fid = AFS_FS_I(inode)->fid;
966 d = d_splice_alias(inode, dentry);
967 if (!IS_ERR_OR_NULL(d)) {
968 d->d_fsdata = dentry->d_fsdata;
969 trace_afs_lookup(dvnode, &d->d_name, &fid);
970 } else {
971 trace_afs_lookup(dvnode, &dentry->d_name, &fid);
973 return d;
977 * Check the validity of a dentry under RCU conditions.
979 static int afs_d_revalidate_rcu(struct dentry *dentry)
981 struct afs_vnode *dvnode, *vnode;
982 struct dentry *parent;
983 struct inode *dir, *inode;
984 long dir_version, de_version;
986 _enter("%p", dentry);
988 /* Check the parent directory is still valid first. */
989 parent = READ_ONCE(dentry->d_parent);
990 dir = d_inode_rcu(parent);
991 if (!dir)
992 return -ECHILD;
993 dvnode = AFS_FS_I(dir);
994 if (test_bit(AFS_VNODE_DELETED, &dvnode->flags))
995 return -ECHILD;
997 if (!afs_check_validity(dvnode))
998 return -ECHILD;
1000 /* We only need to invalidate a dentry if the server's copy changed
1001 * behind our back. If we made the change, it's no problem. Note that
1002 * on a 32-bit system, we only have 32 bits in the dentry to store the
1003 * version.
1005 dir_version = (long)READ_ONCE(dvnode->status.data_version);
1006 de_version = (long)READ_ONCE(dentry->d_fsdata);
1007 if (de_version != dir_version) {
1008 dir_version = (long)READ_ONCE(dvnode->invalid_before);
1009 if (de_version - dir_version < 0)
1010 return -ECHILD;
1013 /* Check to see if the vnode referred to by the dentry still
1014 * has a callback.
1016 if (d_really_is_positive(dentry)) {
1017 inode = d_inode_rcu(dentry);
1018 if (inode) {
1019 vnode = AFS_FS_I(inode);
1020 if (!afs_check_validity(vnode))
1021 return -ECHILD;
1025 return 1; /* Still valid */
1029 * check that a dentry lookup hit has found a valid entry
1030 * - NOTE! the hit can be a negative hit too, so we can't assume we have an
1031 * inode
1033 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags)
1035 struct afs_vnode *vnode, *dir;
1036 struct afs_fid uninitialized_var(fid);
1037 struct dentry *parent;
1038 struct inode *inode;
1039 struct key *key;
1040 afs_dataversion_t dir_version, invalid_before;
1041 long de_version;
1042 int ret;
1044 if (flags & LOOKUP_RCU)
1045 return afs_d_revalidate_rcu(dentry);
1047 if (d_really_is_positive(dentry)) {
1048 vnode = AFS_FS_I(d_inode(dentry));
1049 _enter("{v={%llx:%llu} n=%pd fl=%lx},",
1050 vnode->fid.vid, vnode->fid.vnode, dentry,
1051 vnode->flags);
1052 } else {
1053 _enter("{neg n=%pd}", dentry);
1056 key = afs_request_key(AFS_FS_S(dentry->d_sb)->volume->cell);
1057 if (IS_ERR(key))
1058 key = NULL;
1060 if (d_really_is_positive(dentry)) {
1061 inode = d_inode(dentry);
1062 if (inode) {
1063 vnode = AFS_FS_I(inode);
1064 afs_validate(vnode, key);
1065 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1066 goto out_bad;
1070 /* lock down the parent dentry so we can peer at it */
1071 parent = dget_parent(dentry);
1072 dir = AFS_FS_I(d_inode(parent));
1074 /* validate the parent directory */
1075 afs_validate(dir, key);
1077 if (test_bit(AFS_VNODE_DELETED, &dir->flags)) {
1078 _debug("%pd: parent dir deleted", dentry);
1079 goto out_bad_parent;
1082 /* We only need to invalidate a dentry if the server's copy changed
1083 * behind our back. If we made the change, it's no problem. Note that
1084 * on a 32-bit system, we only have 32 bits in the dentry to store the
1085 * version.
1087 dir_version = dir->status.data_version;
1088 de_version = (long)dentry->d_fsdata;
1089 if (de_version == (long)dir_version)
1090 goto out_valid_noupdate;
1092 invalid_before = dir->invalid_before;
1093 if (de_version - (long)invalid_before >= 0)
1094 goto out_valid;
1096 _debug("dir modified");
1097 afs_stat_v(dir, n_reval);
1099 /* search the directory for this vnode */
1100 ret = afs_do_lookup_one(&dir->vfs_inode, dentry, &fid, key, &dir_version);
1101 switch (ret) {
1102 case 0:
1103 /* the filename maps to something */
1104 if (d_really_is_negative(dentry))
1105 goto out_bad_parent;
1106 inode = d_inode(dentry);
1107 if (is_bad_inode(inode)) {
1108 printk("kAFS: afs_d_revalidate: %pd2 has bad inode\n",
1109 dentry);
1110 goto out_bad_parent;
1113 vnode = AFS_FS_I(inode);
1115 /* if the vnode ID has changed, then the dirent points to a
1116 * different file */
1117 if (fid.vnode != vnode->fid.vnode) {
1118 _debug("%pd: dirent changed [%llu != %llu]",
1119 dentry, fid.vnode,
1120 vnode->fid.vnode);
1121 goto not_found;
1124 /* if the vnode ID uniqifier has changed, then the file has
1125 * been deleted and replaced, and the original vnode ID has
1126 * been reused */
1127 if (fid.unique != vnode->fid.unique) {
1128 _debug("%pd: file deleted (uq %u -> %u I:%u)",
1129 dentry, fid.unique,
1130 vnode->fid.unique,
1131 vnode->vfs_inode.i_generation);
1132 write_seqlock(&vnode->cb_lock);
1133 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1134 write_sequnlock(&vnode->cb_lock);
1135 goto not_found;
1137 goto out_valid;
1139 case -ENOENT:
1140 /* the filename is unknown */
1141 _debug("%pd: dirent not found", dentry);
1142 if (d_really_is_positive(dentry))
1143 goto not_found;
1144 goto out_valid;
1146 default:
1147 _debug("failed to iterate dir %pd: %d",
1148 parent, ret);
1149 goto out_bad_parent;
1152 out_valid:
1153 dentry->d_fsdata = (void *)(unsigned long)dir_version;
1154 out_valid_noupdate:
1155 dput(parent);
1156 key_put(key);
1157 _leave(" = 1 [valid]");
1158 return 1;
1160 /* the dirent, if it exists, now points to a different vnode */
1161 not_found:
1162 spin_lock(&dentry->d_lock);
1163 dentry->d_flags |= DCACHE_NFSFS_RENAMED;
1164 spin_unlock(&dentry->d_lock);
1166 out_bad_parent:
1167 _debug("dropping dentry %pd2", dentry);
1168 dput(parent);
1169 out_bad:
1170 key_put(key);
1172 _leave(" = 0 [bad]");
1173 return 0;
1177 * allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
1178 * sleep)
1179 * - called from dput() when d_count is going to 0.
1180 * - return 1 to request dentry be unhashed, 0 otherwise
1182 static int afs_d_delete(const struct dentry *dentry)
1184 _enter("%pd", dentry);
1186 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1187 goto zap;
1189 if (d_really_is_positive(dentry) &&
1190 (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(d_inode(dentry))->flags) ||
1191 test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(d_inode(dentry))->flags)))
1192 goto zap;
1194 _leave(" = 0 [keep]");
1195 return 0;
1197 zap:
1198 _leave(" = 1 [zap]");
1199 return 1;
1203 * Clean up sillyrename files on dentry removal.
1205 static void afs_d_iput(struct dentry *dentry, struct inode *inode)
1207 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1208 afs_silly_iput(dentry, inode);
1209 iput(inode);
1213 * handle dentry release
1215 void afs_d_release(struct dentry *dentry)
1217 _enter("%pd", dentry);
1221 * Create a new inode for create/mkdir/symlink
1223 static void afs_vnode_new_inode(struct afs_fs_cursor *fc,
1224 struct dentry *new_dentry,
1225 struct afs_iget_data *new_data,
1226 struct afs_status_cb *new_scb)
1228 struct afs_vnode *vnode;
1229 struct inode *inode;
1231 if (fc->ac.error < 0)
1232 return;
1234 inode = afs_iget(fc->vnode->vfs_inode.i_sb, fc->key,
1235 new_data, new_scb, fc->cbi, fc->vnode);
1236 if (IS_ERR(inode)) {
1237 /* ENOMEM or EINTR at a really inconvenient time - just abandon
1238 * the new directory on the server.
1240 fc->ac.error = PTR_ERR(inode);
1241 return;
1244 vnode = AFS_FS_I(inode);
1245 set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
1246 if (fc->ac.error == 0)
1247 afs_cache_permit(vnode, fc->key, vnode->cb_break, new_scb);
1248 d_instantiate(new_dentry, inode);
1251 static void afs_prep_for_new_inode(struct afs_fs_cursor *fc,
1252 struct afs_iget_data *iget_data)
1254 iget_data->volume = fc->vnode->volume;
1255 iget_data->cb_v_break = fc->vnode->volume->cb_v_break;
1256 iget_data->cb_s_break = fc->cbi->server->cb_s_break;
1260 * Note that a dentry got changed. We need to set d_fsdata to the data version
1261 * number derived from the result of the operation. It doesn't matter if
1262 * d_fsdata goes backwards as we'll just revalidate.
1264 static void afs_update_dentry_version(struct afs_fs_cursor *fc,
1265 struct dentry *dentry,
1266 struct afs_status_cb *scb)
1268 if (fc->ac.error == 0)
1269 dentry->d_fsdata =
1270 (void *)(unsigned long)scb->status.data_version;
1274 * create a directory on an AFS filesystem
1276 static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1278 struct afs_iget_data iget_data;
1279 struct afs_status_cb *scb;
1280 struct afs_fs_cursor fc;
1281 struct afs_vnode *dvnode = AFS_FS_I(dir);
1282 struct key *key;
1283 afs_dataversion_t data_version;
1284 int ret;
1286 mode |= S_IFDIR;
1288 _enter("{%llx:%llu},{%pd},%ho",
1289 dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1291 ret = -ENOMEM;
1292 scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1293 if (!scb)
1294 goto error;
1296 key = afs_request_key(dvnode->volume->cell);
1297 if (IS_ERR(key)) {
1298 ret = PTR_ERR(key);
1299 goto error_scb;
1302 ret = -ERESTARTSYS;
1303 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1304 data_version = dvnode->status.data_version + 1;
1306 while (afs_select_fileserver(&fc)) {
1307 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1308 afs_prep_for_new_inode(&fc, &iget_data);
1309 afs_fs_create(&fc, dentry->d_name.name, mode,
1310 &scb[0], &iget_data.fid, &scb[1]);
1313 afs_check_for_remote_deletion(&fc, dvnode);
1314 afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1315 &data_version, &scb[0]);
1316 afs_update_dentry_version(&fc, dentry, &scb[0]);
1317 afs_vnode_new_inode(&fc, dentry, &iget_data, &scb[1]);
1318 ret = afs_end_vnode_operation(&fc);
1319 if (ret < 0)
1320 goto error_key;
1321 } else {
1322 goto error_key;
1325 if (ret == 0) {
1326 down_write(&dvnode->validate_lock);
1327 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1328 dvnode->status.data_version == data_version)
1329 afs_edit_dir_add(dvnode, &dentry->d_name, &iget_data.fid,
1330 afs_edit_dir_for_create);
1331 up_write(&dvnode->validate_lock);
1334 key_put(key);
1335 kfree(scb);
1336 _leave(" = 0");
1337 return 0;
1339 error_key:
1340 key_put(key);
1341 error_scb:
1342 kfree(scb);
1343 error:
1344 d_drop(dentry);
1345 _leave(" = %d", ret);
1346 return ret;
1350 * Remove a subdir from a directory.
1352 static void afs_dir_remove_subdir(struct dentry *dentry)
1354 if (d_really_is_positive(dentry)) {
1355 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1357 clear_nlink(&vnode->vfs_inode);
1358 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1359 clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1360 clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
1365 * remove a directory from an AFS filesystem
1367 static int afs_rmdir(struct inode *dir, struct dentry *dentry)
1369 struct afs_status_cb *scb;
1370 struct afs_fs_cursor fc;
1371 struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode = NULL;
1372 struct key *key;
1373 afs_dataversion_t data_version;
1374 int ret;
1376 _enter("{%llx:%llu},{%pd}",
1377 dvnode->fid.vid, dvnode->fid.vnode, dentry);
1379 scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL);
1380 if (!scb)
1381 return -ENOMEM;
1383 key = afs_request_key(dvnode->volume->cell);
1384 if (IS_ERR(key)) {
1385 ret = PTR_ERR(key);
1386 goto error;
1389 /* Try to make sure we have a callback promise on the victim. */
1390 if (d_really_is_positive(dentry)) {
1391 vnode = AFS_FS_I(d_inode(dentry));
1392 ret = afs_validate(vnode, key);
1393 if (ret < 0)
1394 goto error_key;
1397 if (vnode) {
1398 ret = down_write_killable(&vnode->rmdir_lock);
1399 if (ret < 0)
1400 goto error_key;
1403 ret = -ERESTARTSYS;
1404 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1405 data_version = dvnode->status.data_version + 1;
1407 while (afs_select_fileserver(&fc)) {
1408 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1409 afs_fs_remove(&fc, vnode, dentry->d_name.name, true, scb);
1412 afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1413 &data_version, scb);
1414 afs_update_dentry_version(&fc, dentry, scb);
1415 ret = afs_end_vnode_operation(&fc);
1416 if (ret == 0) {
1417 afs_dir_remove_subdir(dentry);
1418 down_write(&dvnode->validate_lock);
1419 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1420 dvnode->status.data_version == data_version)
1421 afs_edit_dir_remove(dvnode, &dentry->d_name,
1422 afs_edit_dir_for_rmdir);
1423 up_write(&dvnode->validate_lock);
1427 if (vnode)
1428 up_write(&vnode->rmdir_lock);
1429 error_key:
1430 key_put(key);
1431 error:
1432 kfree(scb);
1433 return ret;
1437 * Remove a link to a file or symlink from a directory.
1439 * If the file was not deleted due to excess hard links, the fileserver will
1440 * break the callback promise on the file - if it had one - before it returns
1441 * to us, and if it was deleted, it won't
1443 * However, if we didn't have a callback promise outstanding, or it was
1444 * outstanding on a different server, then it won't break it either...
1446 static int afs_dir_remove_link(struct afs_vnode *dvnode, struct dentry *dentry,
1447 struct key *key)
1449 int ret = 0;
1451 if (d_really_is_positive(dentry)) {
1452 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1454 if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
1455 /* Already done */
1456 } else if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
1457 write_seqlock(&vnode->cb_lock);
1458 drop_nlink(&vnode->vfs_inode);
1459 if (vnode->vfs_inode.i_nlink == 0) {
1460 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1461 __afs_break_callback(vnode, afs_cb_break_for_unlink);
1463 write_sequnlock(&vnode->cb_lock);
1464 ret = 0;
1465 } else {
1466 afs_break_callback(vnode, afs_cb_break_for_unlink);
1468 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1469 kdebug("AFS_VNODE_DELETED");
1471 ret = afs_validate(vnode, key);
1472 if (ret == -ESTALE)
1473 ret = 0;
1475 _debug("nlink %d [val %d]", vnode->vfs_inode.i_nlink, ret);
1478 return ret;
1482 * Remove a file or symlink from an AFS filesystem.
1484 static int afs_unlink(struct inode *dir, struct dentry *dentry)
1486 struct afs_fs_cursor fc;
1487 struct afs_status_cb *scb;
1488 struct afs_vnode *dvnode = AFS_FS_I(dir);
1489 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1490 struct key *key;
1491 bool need_rehash = false;
1492 int ret;
1494 _enter("{%llx:%llu},{%pd}",
1495 dvnode->fid.vid, dvnode->fid.vnode, dentry);
1497 if (dentry->d_name.len >= AFSNAMEMAX)
1498 return -ENAMETOOLONG;
1500 ret = -ENOMEM;
1501 scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1502 if (!scb)
1503 goto error;
1505 key = afs_request_key(dvnode->volume->cell);
1506 if (IS_ERR(key)) {
1507 ret = PTR_ERR(key);
1508 goto error_scb;
1511 /* Try to make sure we have a callback promise on the victim. */
1512 ret = afs_validate(vnode, key);
1513 if (ret < 0)
1514 goto error_key;
1516 spin_lock(&dentry->d_lock);
1517 if (d_count(dentry) > 1) {
1518 spin_unlock(&dentry->d_lock);
1519 /* Start asynchronous writeout of the inode */
1520 write_inode_now(d_inode(dentry), 0);
1521 ret = afs_sillyrename(dvnode, vnode, dentry, key);
1522 goto error_key;
1524 if (!d_unhashed(dentry)) {
1525 /* Prevent a race with RCU lookup. */
1526 __d_drop(dentry);
1527 need_rehash = true;
1529 spin_unlock(&dentry->d_lock);
1531 ret = -ERESTARTSYS;
1532 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1533 afs_dataversion_t data_version = dvnode->status.data_version + 1;
1534 afs_dataversion_t data_version_2 = vnode->status.data_version;
1536 while (afs_select_fileserver(&fc)) {
1537 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1538 fc.cb_break_2 = afs_calc_vnode_cb_break(vnode);
1540 if (test_bit(AFS_SERVER_FL_IS_YFS, &fc.cbi->server->flags) &&
1541 !test_bit(AFS_SERVER_FL_NO_RM2, &fc.cbi->server->flags)) {
1542 yfs_fs_remove_file2(&fc, vnode, dentry->d_name.name,
1543 &scb[0], &scb[1]);
1544 if (fc.ac.error != -ECONNABORTED ||
1545 fc.ac.abort_code != RXGEN_OPCODE)
1546 continue;
1547 set_bit(AFS_SERVER_FL_NO_RM2, &fc.cbi->server->flags);
1550 afs_fs_remove(&fc, vnode, dentry->d_name.name, false, &scb[0]);
1553 afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1554 &data_version, &scb[0]);
1555 afs_vnode_commit_status(&fc, vnode, fc.cb_break_2,
1556 &data_version_2, &scb[1]);
1557 afs_update_dentry_version(&fc, dentry, &scb[0]);
1558 ret = afs_end_vnode_operation(&fc);
1559 if (ret == 0 && !(scb[1].have_status || scb[1].have_error))
1560 ret = afs_dir_remove_link(dvnode, dentry, key);
1562 if (ret == 0) {
1563 down_write(&dvnode->validate_lock);
1564 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1565 dvnode->status.data_version == data_version)
1566 afs_edit_dir_remove(dvnode, &dentry->d_name,
1567 afs_edit_dir_for_unlink);
1568 up_write(&dvnode->validate_lock);
1572 if (need_rehash && ret < 0 && ret != -ENOENT)
1573 d_rehash(dentry);
1575 error_key:
1576 key_put(key);
1577 error_scb:
1578 kfree(scb);
1579 error:
1580 _leave(" = %d", ret);
1581 return ret;
1585 * create a regular file on an AFS filesystem
1587 static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1588 bool excl)
1590 struct afs_iget_data iget_data;
1591 struct afs_fs_cursor fc;
1592 struct afs_status_cb *scb;
1593 struct afs_vnode *dvnode = AFS_FS_I(dir);
1594 struct key *key;
1595 afs_dataversion_t data_version;
1596 int ret;
1598 mode |= S_IFREG;
1600 _enter("{%llx:%llu},{%pd},%ho,",
1601 dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1603 ret = -ENAMETOOLONG;
1604 if (dentry->d_name.len >= AFSNAMEMAX)
1605 goto error;
1607 key = afs_request_key(dvnode->volume->cell);
1608 if (IS_ERR(key)) {
1609 ret = PTR_ERR(key);
1610 goto error;
1613 ret = -ENOMEM;
1614 scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1615 if (!scb)
1616 goto error_scb;
1618 ret = -ERESTARTSYS;
1619 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1620 data_version = dvnode->status.data_version + 1;
1622 while (afs_select_fileserver(&fc)) {
1623 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1624 afs_prep_for_new_inode(&fc, &iget_data);
1625 afs_fs_create(&fc, dentry->d_name.name, mode,
1626 &scb[0], &iget_data.fid, &scb[1]);
1629 afs_check_for_remote_deletion(&fc, dvnode);
1630 afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1631 &data_version, &scb[0]);
1632 afs_update_dentry_version(&fc, dentry, &scb[0]);
1633 afs_vnode_new_inode(&fc, dentry, &iget_data, &scb[1]);
1634 ret = afs_end_vnode_operation(&fc);
1635 if (ret < 0)
1636 goto error_key;
1637 } else {
1638 goto error_key;
1641 down_write(&dvnode->validate_lock);
1642 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1643 dvnode->status.data_version == data_version)
1644 afs_edit_dir_add(dvnode, &dentry->d_name, &iget_data.fid,
1645 afs_edit_dir_for_create);
1646 up_write(&dvnode->validate_lock);
1648 kfree(scb);
1649 key_put(key);
1650 _leave(" = 0");
1651 return 0;
1653 error_scb:
1654 kfree(scb);
1655 error_key:
1656 key_put(key);
1657 error:
1658 d_drop(dentry);
1659 _leave(" = %d", ret);
1660 return ret;
1664 * create a hard link between files in an AFS filesystem
1666 static int afs_link(struct dentry *from, struct inode *dir,
1667 struct dentry *dentry)
1669 struct afs_fs_cursor fc;
1670 struct afs_status_cb *scb;
1671 struct afs_vnode *dvnode = AFS_FS_I(dir);
1672 struct afs_vnode *vnode = AFS_FS_I(d_inode(from));
1673 struct key *key;
1674 afs_dataversion_t data_version;
1675 int ret;
1677 _enter("{%llx:%llu},{%llx:%llu},{%pd}",
1678 vnode->fid.vid, vnode->fid.vnode,
1679 dvnode->fid.vid, dvnode->fid.vnode,
1680 dentry);
1682 ret = -ENAMETOOLONG;
1683 if (dentry->d_name.len >= AFSNAMEMAX)
1684 goto error;
1686 ret = -ENOMEM;
1687 scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1688 if (!scb)
1689 goto error;
1691 key = afs_request_key(dvnode->volume->cell);
1692 if (IS_ERR(key)) {
1693 ret = PTR_ERR(key);
1694 goto error_scb;
1697 ret = -ERESTARTSYS;
1698 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1699 data_version = dvnode->status.data_version + 1;
1701 if (mutex_lock_interruptible_nested(&vnode->io_lock, 1) < 0) {
1702 afs_end_vnode_operation(&fc);
1703 goto error_key;
1706 while (afs_select_fileserver(&fc)) {
1707 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1708 fc.cb_break_2 = afs_calc_vnode_cb_break(vnode);
1709 afs_fs_link(&fc, vnode, dentry->d_name.name,
1710 &scb[0], &scb[1]);
1713 afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1714 &data_version, &scb[0]);
1715 afs_vnode_commit_status(&fc, vnode, fc.cb_break_2,
1716 NULL, &scb[1]);
1717 ihold(&vnode->vfs_inode);
1718 afs_update_dentry_version(&fc, dentry, &scb[0]);
1719 d_instantiate(dentry, &vnode->vfs_inode);
1721 mutex_unlock(&vnode->io_lock);
1722 ret = afs_end_vnode_operation(&fc);
1723 if (ret < 0)
1724 goto error_key;
1725 } else {
1726 goto error_key;
1729 down_write(&dvnode->validate_lock);
1730 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1731 dvnode->status.data_version == data_version)
1732 afs_edit_dir_add(dvnode, &dentry->d_name, &vnode->fid,
1733 afs_edit_dir_for_link);
1734 up_write(&dvnode->validate_lock);
1736 key_put(key);
1737 kfree(scb);
1738 _leave(" = 0");
1739 return 0;
1741 error_key:
1742 key_put(key);
1743 error_scb:
1744 kfree(scb);
1745 error:
1746 d_drop(dentry);
1747 _leave(" = %d", ret);
1748 return ret;
1752 * create a symlink in an AFS filesystem
1754 static int afs_symlink(struct inode *dir, struct dentry *dentry,
1755 const char *content)
1757 struct afs_iget_data iget_data;
1758 struct afs_fs_cursor fc;
1759 struct afs_status_cb *scb;
1760 struct afs_vnode *dvnode = AFS_FS_I(dir);
1761 struct key *key;
1762 afs_dataversion_t data_version;
1763 int ret;
1765 _enter("{%llx:%llu},{%pd},%s",
1766 dvnode->fid.vid, dvnode->fid.vnode, dentry,
1767 content);
1769 ret = -ENAMETOOLONG;
1770 if (dentry->d_name.len >= AFSNAMEMAX)
1771 goto error;
1773 ret = -EINVAL;
1774 if (strlen(content) >= AFSPATHMAX)
1775 goto error;
1777 ret = -ENOMEM;
1778 scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1779 if (!scb)
1780 goto error;
1782 key = afs_request_key(dvnode->volume->cell);
1783 if (IS_ERR(key)) {
1784 ret = PTR_ERR(key);
1785 goto error_scb;
1788 ret = -ERESTARTSYS;
1789 if (afs_begin_vnode_operation(&fc, dvnode, key, true)) {
1790 data_version = dvnode->status.data_version + 1;
1792 while (afs_select_fileserver(&fc)) {
1793 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1794 afs_prep_for_new_inode(&fc, &iget_data);
1795 afs_fs_symlink(&fc, dentry->d_name.name, content,
1796 &scb[0], &iget_data.fid, &scb[1]);
1799 afs_check_for_remote_deletion(&fc, dvnode);
1800 afs_vnode_commit_status(&fc, dvnode, fc.cb_break,
1801 &data_version, &scb[0]);
1802 afs_update_dentry_version(&fc, dentry, &scb[0]);
1803 afs_vnode_new_inode(&fc, dentry, &iget_data, &scb[1]);
1804 ret = afs_end_vnode_operation(&fc);
1805 if (ret < 0)
1806 goto error_key;
1807 } else {
1808 goto error_key;
1811 down_write(&dvnode->validate_lock);
1812 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1813 dvnode->status.data_version == data_version)
1814 afs_edit_dir_add(dvnode, &dentry->d_name, &iget_data.fid,
1815 afs_edit_dir_for_symlink);
1816 up_write(&dvnode->validate_lock);
1818 key_put(key);
1819 kfree(scb);
1820 _leave(" = 0");
1821 return 0;
1823 error_key:
1824 key_put(key);
1825 error_scb:
1826 kfree(scb);
1827 error:
1828 d_drop(dentry);
1829 _leave(" = %d", ret);
1830 return ret;
1834 * rename a file in an AFS filesystem and/or move it between directories
1836 static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
1837 struct inode *new_dir, struct dentry *new_dentry,
1838 unsigned int flags)
1840 struct afs_fs_cursor fc;
1841 struct afs_status_cb *scb;
1842 struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
1843 struct dentry *tmp = NULL, *rehash = NULL;
1844 struct inode *new_inode;
1845 struct key *key;
1846 afs_dataversion_t orig_data_version;
1847 afs_dataversion_t new_data_version;
1848 bool new_negative = d_is_negative(new_dentry);
1849 int ret;
1851 if (flags)
1852 return -EINVAL;
1854 /* Don't allow silly-rename files be moved around. */
1855 if (old_dentry->d_flags & DCACHE_NFSFS_RENAMED)
1856 return -EINVAL;
1858 vnode = AFS_FS_I(d_inode(old_dentry));
1859 orig_dvnode = AFS_FS_I(old_dir);
1860 new_dvnode = AFS_FS_I(new_dir);
1862 _enter("{%llx:%llu},{%llx:%llu},{%llx:%llu},{%pd}",
1863 orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
1864 vnode->fid.vid, vnode->fid.vnode,
1865 new_dvnode->fid.vid, new_dvnode->fid.vnode,
1866 new_dentry);
1868 ret = -ENOMEM;
1869 scb = kcalloc(2, sizeof(struct afs_status_cb), GFP_KERNEL);
1870 if (!scb)
1871 goto error;
1873 key = afs_request_key(orig_dvnode->volume->cell);
1874 if (IS_ERR(key)) {
1875 ret = PTR_ERR(key);
1876 goto error_scb;
1879 /* For non-directories, check whether the target is busy and if so,
1880 * make a copy of the dentry and then do a silly-rename. If the
1881 * silly-rename succeeds, the copied dentry is hashed and becomes the
1882 * new target.
1884 if (d_is_positive(new_dentry) && !d_is_dir(new_dentry)) {
1885 /* To prevent any new references to the target during the
1886 * rename, we unhash the dentry in advance.
1888 if (!d_unhashed(new_dentry)) {
1889 d_drop(new_dentry);
1890 rehash = new_dentry;
1893 if (d_count(new_dentry) > 2) {
1894 /* copy the target dentry's name */
1895 ret = -ENOMEM;
1896 tmp = d_alloc(new_dentry->d_parent,
1897 &new_dentry->d_name);
1898 if (!tmp)
1899 goto error_rehash;
1901 ret = afs_sillyrename(new_dvnode,
1902 AFS_FS_I(d_inode(new_dentry)),
1903 new_dentry, key);
1904 if (ret)
1905 goto error_rehash;
1907 new_dentry = tmp;
1908 rehash = NULL;
1909 new_negative = true;
1913 /* This bit is potentially nasty as there's a potential race with
1914 * afs_d_revalidate{,_rcu}(). We have to change d_fsdata on the dentry
1915 * to reflect it's new parent's new data_version after the op, but
1916 * d_revalidate may see old_dentry between the op having taken place
1917 * and the version being updated.
1919 * So drop the old_dentry for now to make other threads go through
1920 * lookup instead - which we hold a lock against.
1922 d_drop(old_dentry);
1924 ret = -ERESTARTSYS;
1925 if (afs_begin_vnode_operation(&fc, orig_dvnode, key, true)) {
1926 orig_data_version = orig_dvnode->status.data_version + 1;
1928 if (orig_dvnode != new_dvnode) {
1929 if (mutex_lock_interruptible_nested(&new_dvnode->io_lock, 1) < 0) {
1930 afs_end_vnode_operation(&fc);
1931 goto error_rehash_old;
1933 new_data_version = new_dvnode->status.data_version + 1;
1934 } else {
1935 new_data_version = orig_data_version;
1938 while (afs_select_fileserver(&fc)) {
1939 fc.cb_break = afs_calc_vnode_cb_break(orig_dvnode);
1940 fc.cb_break_2 = afs_calc_vnode_cb_break(new_dvnode);
1941 afs_fs_rename(&fc, old_dentry->d_name.name,
1942 new_dvnode, new_dentry->d_name.name,
1943 &scb[0], &scb[1]);
1946 afs_vnode_commit_status(&fc, orig_dvnode, fc.cb_break,
1947 &orig_data_version, &scb[0]);
1948 if (new_dvnode != orig_dvnode) {
1949 afs_vnode_commit_status(&fc, new_dvnode, fc.cb_break_2,
1950 &new_data_version, &scb[1]);
1951 mutex_unlock(&new_dvnode->io_lock);
1953 ret = afs_end_vnode_operation(&fc);
1954 if (ret < 0)
1955 goto error_rehash_old;
1958 if (ret == 0) {
1959 if (rehash)
1960 d_rehash(rehash);
1961 down_write(&orig_dvnode->validate_lock);
1962 if (test_bit(AFS_VNODE_DIR_VALID, &orig_dvnode->flags) &&
1963 orig_dvnode->status.data_version == orig_data_version)
1964 afs_edit_dir_remove(orig_dvnode, &old_dentry->d_name,
1965 afs_edit_dir_for_rename_0);
1966 if (orig_dvnode != new_dvnode) {
1967 up_write(&orig_dvnode->validate_lock);
1969 down_write(&new_dvnode->validate_lock);
1971 if (test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags) &&
1972 orig_dvnode->status.data_version == new_data_version) {
1973 if (!new_negative)
1974 afs_edit_dir_remove(new_dvnode, &new_dentry->d_name,
1975 afs_edit_dir_for_rename_1);
1977 afs_edit_dir_add(new_dvnode, &new_dentry->d_name,
1978 &vnode->fid, afs_edit_dir_for_rename_2);
1981 new_inode = d_inode(new_dentry);
1982 if (new_inode) {
1983 spin_lock(&new_inode->i_lock);
1984 if (new_inode->i_nlink > 0)
1985 drop_nlink(new_inode);
1986 spin_unlock(&new_inode->i_lock);
1989 /* Now we can update d_fsdata on the dentries to reflect their
1990 * new parent's data_version.
1992 * Note that if we ever implement RENAME_EXCHANGE, we'll have
1993 * to update both dentries with opposing dir versions.
1995 afs_update_dentry_version(&fc, old_dentry, &scb[1]);
1996 afs_update_dentry_version(&fc, new_dentry, &scb[1]);
1997 d_move(old_dentry, new_dentry);
1998 up_write(&new_dvnode->validate_lock);
1999 goto error_tmp;
2002 error_rehash_old:
2003 d_rehash(new_dentry);
2004 error_rehash:
2005 if (rehash)
2006 d_rehash(rehash);
2007 error_tmp:
2008 if (tmp)
2009 dput(tmp);
2010 key_put(key);
2011 error_scb:
2012 kfree(scb);
2013 error:
2014 _leave(" = %d", ret);
2015 return ret;
2019 * Release a directory page and clean up its private state if it's not busy
2020 * - return true if the page can now be released, false if not
2022 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags)
2024 struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
2026 _enter("{{%llx:%llu}[%lu]}", dvnode->fid.vid, dvnode->fid.vnode, page->index);
2028 set_page_private(page, 0);
2029 ClearPagePrivate(page);
2031 /* The directory will need reloading. */
2032 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2033 afs_stat_v(dvnode, n_relpg);
2034 return 1;
2038 * invalidate part or all of a page
2039 * - release a page and clean up its private data if offset is 0 (indicating
2040 * the entire page)
2042 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
2043 unsigned int length)
2045 struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
2047 _enter("{%lu},%u,%u", page->index, offset, length);
2049 BUG_ON(!PageLocked(page));
2051 /* The directory will need reloading. */
2052 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2053 afs_stat_v(dvnode, n_inval);
2055 /* we clean up only if the entire page is being invalidated */
2056 if (offset == 0 && length == PAGE_SIZE) {
2057 set_page_private(page, 0);
2058 ClearPagePrivate(page);