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)
8 #include <linux/kernel.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>
20 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
22 static int afs_dir_open(struct inode *inode, struct file *file);
23 static int afs_readdir(struct file *file, struct dir_context *ctx);
24 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags);
25 static int afs_d_delete(const struct dentry *dentry);
26 static void afs_d_iput(struct dentry *dentry, struct inode *inode);
27 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name, int nlen,
28 loff_t fpos, u64 ino, unsigned dtype);
29 static int afs_lookup_filldir(struct dir_context *ctx, const char *name, int nlen,
30 loff_t fpos, u64 ino, unsigned dtype);
31 static int afs_create(struct user_namespace *mnt_userns, struct inode *dir,
32 struct dentry *dentry, umode_t mode, bool excl);
33 static int afs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
34 struct dentry *dentry, umode_t mode);
35 static int afs_rmdir(struct inode *dir, struct dentry *dentry);
36 static int afs_unlink(struct inode *dir, struct dentry *dentry);
37 static int afs_link(struct dentry *from, struct inode *dir,
38 struct dentry *dentry);
39 static int afs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
40 struct dentry *dentry, const char *content);
41 static int afs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
42 struct dentry *old_dentry, struct inode *new_dir,
43 struct dentry *new_dentry, unsigned int flags);
44 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags);
45 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
48 static int afs_dir_set_page_dirty(struct page *page)
50 BUG(); /* This should never happen. */
53 const struct file_operations afs_dir_file_operations = {
55 .release = afs_release,
56 .iterate_shared = afs_readdir,
58 .llseek = generic_file_llseek,
61 const struct inode_operations afs_dir_inode_operations = {
66 .symlink = afs_symlink,
70 .permission = afs_permission,
71 .getattr = afs_getattr,
72 .setattr = afs_setattr,
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,
89 struct afs_lookup_one_cookie {
90 struct dir_context ctx;
96 struct afs_lookup_cookie {
97 struct dir_context ctx;
101 unsigned short nr_fids;
102 struct afs_fid fids[50];
106 * Drop the refs that we're holding on the pages we were reading into. We've
107 * got refs on the first nr_pages pages.
109 static void afs_dir_read_cleanup(struct afs_read *req)
111 struct address_space *mapping = req->vnode->vfs_inode.i_mapping;
113 pgoff_t last = req->nr_pages - 1;
115 XA_STATE(xas, &mapping->i_pages, 0);
117 if (unlikely(!req->nr_pages))
121 xas_for_each(&xas, page, last) {
122 if (xas_retry(&xas, page))
124 BUG_ON(xa_is_value(page));
125 BUG_ON(PageCompound(page));
126 ASSERTCMP(page->mapping, ==, mapping);
135 * check that a directory page is valid
137 static bool afs_dir_check_page(struct afs_vnode *dvnode, struct page *page,
140 struct afs_xdr_dir_page *dbuf;
144 /* Determine how many magic numbers there should be in this page, but
145 * we must take care because the directory may change size under us.
147 off = page_offset(page);
151 latter = i_size - off;
152 if (latter >= PAGE_SIZE)
156 qty /= sizeof(union afs_xdr_dir_block);
159 dbuf = kmap_atomic(page);
160 for (tmp = 0; tmp < qty; tmp++) {
161 if (dbuf->blocks[tmp].hdr.magic != AFS_DIR_MAGIC) {
162 printk("kAFS: %s(%lx): bad magic %d/%d is %04hx\n",
163 __func__, dvnode->vfs_inode.i_ino, tmp, qty,
164 ntohs(dbuf->blocks[tmp].hdr.magic));
165 trace_afs_dir_check_failed(dvnode, off, i_size);
167 trace_afs_file_error(dvnode, -EIO, afs_file_error_dir_bad_magic);
171 /* Make sure each block is NUL terminated so we can reasonably
172 * use string functions on it. The filenames in the page
173 * *should* be NUL-terminated anyway.
175 ((u8 *)&dbuf->blocks[tmp])[AFS_DIR_BLOCK_SIZE - 1] = 0;
181 afs_stat_v(dvnode, n_read_dir);
189 * Dump the contents of a directory.
191 static void afs_dir_dump(struct afs_vnode *dvnode, struct afs_read *req)
193 struct afs_xdr_dir_page *dbuf;
194 struct address_space *mapping = dvnode->vfs_inode.i_mapping;
196 unsigned int i, qty = PAGE_SIZE / sizeof(union afs_xdr_dir_block);
197 pgoff_t last = req->nr_pages - 1;
199 XA_STATE(xas, &mapping->i_pages, 0);
201 pr_warn("DIR %llx:%llx f=%llx l=%llx al=%llx\n",
202 dvnode->fid.vid, dvnode->fid.vnode,
203 req->file_size, req->len, req->actual_len);
204 pr_warn("DIR %llx %x %zx %zx\n",
205 req->pos, req->nr_pages,
206 req->iter->iov_offset, iov_iter_count(req->iter));
208 xas_for_each(&xas, page, last) {
209 if (xas_retry(&xas, page))
212 BUG_ON(PageCompound(page));
213 BUG_ON(page->mapping != mapping);
215 dbuf = kmap_atomic(page);
216 for (i = 0; i < qty; i++) {
217 union afs_xdr_dir_block *block = &dbuf->blocks[i];
219 pr_warn("[%02lx] %32phN\n", page->index * qty + i, block);
226 * Check all the pages in a directory. All the pages are held pinned.
228 static int afs_dir_check(struct afs_vnode *dvnode, struct afs_read *req)
230 struct address_space *mapping = dvnode->vfs_inode.i_mapping;
232 pgoff_t last = req->nr_pages - 1;
235 XA_STATE(xas, &mapping->i_pages, 0);
237 if (unlikely(!req->nr_pages))
241 xas_for_each(&xas, page, last) {
242 if (xas_retry(&xas, page))
245 BUG_ON(PageCompound(page));
246 BUG_ON(page->mapping != mapping);
248 if (!afs_dir_check_page(dvnode, page, req->file_size)) {
249 afs_dir_dump(dvnode, req);
260 * open an AFS directory file
262 static int afs_dir_open(struct inode *inode, struct file *file)
264 _enter("{%lu}", inode->i_ino);
266 BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
267 BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
269 if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(inode)->flags))
272 return afs_open(inode, file);
276 * Read the directory into the pagecache in one go, scrubbing the previous
277 * contents. The list of pages is returned, pinning them so that they don't
278 * get reclaimed during the iteration.
280 static struct afs_read *afs_read_dir(struct afs_vnode *dvnode, struct key *key)
281 __acquires(&dvnode->validate_lock)
283 struct afs_read *req;
290 req = kzalloc(sizeof(*req), GFP_KERNEL);
292 return ERR_PTR(-ENOMEM);
294 refcount_set(&req->usage, 1);
296 req->key = key_get(key);
297 req->cleanup = afs_dir_read_cleanup;
300 i_size = i_size_read(&dvnode->vfs_inode);
302 ret = afs_bad(dvnode, afs_file_error_dir_small);
305 if (i_size > 2048 * 1024) {
306 trace_afs_file_error(dvnode, -EFBIG, afs_file_error_dir_big);
311 _enter("%llu", i_size);
313 nr_pages = (i_size + PAGE_SIZE - 1) / PAGE_SIZE;
315 req->actual_len = i_size; /* May change */
316 req->len = nr_pages * PAGE_SIZE; /* We can ask for more than there is */
317 req->data_version = dvnode->status.data_version; /* May change */
318 iov_iter_xarray(&req->def_iter, READ, &dvnode->vfs_inode.i_mapping->i_pages,
320 req->iter = &req->def_iter;
322 /* Fill in any gaps that we might find where the memory reclaimer has
323 * been at work and pin all the pages. If there are any gaps, we will
324 * need to reread the entire directory contents.
327 while (i < nr_pages) {
328 struct page *pages[8], *page;
330 n = find_get_pages_contig(dvnode->vfs_inode.i_mapping, i,
331 min_t(unsigned int, nr_pages - i,
334 _debug("find %u at %u/%u", n, i, nr_pages);
337 gfp_t gfp = dvnode->vfs_inode.i_mapping->gfp_mask;
339 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
340 afs_stat_v(dvnode, n_inval);
343 page = __page_cache_alloc(gfp);
346 ret = add_to_page_cache_lru(page,
347 dvnode->vfs_inode.i_mapping,
352 attach_page_private(page, (void *)1);
362 /* If we're going to reload, we need to lock all the pages to prevent
366 if (down_read_killable(&dvnode->validate_lock) < 0)
369 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
372 up_read(&dvnode->validate_lock);
373 if (down_write_killable(&dvnode->validate_lock) < 0)
376 if (!test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
377 trace_afs_reload_dir(dvnode);
378 ret = afs_fetch_data(dvnode, req);
382 task_io_account_read(PAGE_SIZE * req->nr_pages);
384 if (req->len < req->file_size) {
385 /* The content has grown, so we need to expand the
388 up_write(&dvnode->validate_lock);
392 /* Validate the data we just read. */
393 ret = afs_dir_check(dvnode, req);
397 // TODO: Trim excess pages
399 set_bit(AFS_VNODE_DIR_VALID, &dvnode->flags);
402 downgrade_write(&dvnode->validate_lock);
407 up_write(&dvnode->validate_lock);
410 _leave(" = %d", ret);
415 * deal with one block in an AFS directory
417 static int afs_dir_iterate_block(struct afs_vnode *dvnode,
418 struct dir_context *ctx,
419 union afs_xdr_dir_block *block,
422 union afs_xdr_dirent *dire;
423 unsigned offset, next, curr, nr_slots;
427 _enter("%u,%x,%p,,",(unsigned)ctx->pos,blkoff,block);
429 curr = (ctx->pos - blkoff) / sizeof(union afs_xdr_dirent);
431 /* walk through the block, an entry at a time */
432 for (offset = (blkoff == 0 ? AFS_DIR_RESV_BLOCKS0 : AFS_DIR_RESV_BLOCKS);
433 offset < AFS_DIR_SLOTS_PER_BLOCK;
436 /* skip entries marked unused in the bitmap */
437 if (!(block->hdr.bitmap[offset / 8] &
438 (1 << (offset % 8)))) {
439 _debug("ENT[%zu.%u]: unused",
440 blkoff / sizeof(union afs_xdr_dir_block), offset);
444 next * sizeof(union afs_xdr_dirent);
448 /* got a valid entry */
449 dire = &block->dirents[offset];
450 nlen = strnlen(dire->u.name,
452 offset * sizeof(union afs_xdr_dirent));
453 if (nlen > AFSNAMEMAX - 1) {
454 _debug("ENT[%zu]: name too long (len %u/%zu)",
455 blkoff / sizeof(union afs_xdr_dir_block),
457 return afs_bad(dvnode, afs_file_error_dir_name_too_long);
460 _debug("ENT[%zu.%u]: %s %zu \"%s\"",
461 blkoff / sizeof(union afs_xdr_dir_block), offset,
462 (offset < curr ? "skip" : "fill"),
465 nr_slots = afs_dir_calc_slots(nlen);
466 next = offset + nr_slots;
467 if (next > AFS_DIR_SLOTS_PER_BLOCK) {
468 _debug("ENT[%zu.%u]:"
469 " %u extends beyond end dir block"
471 blkoff / sizeof(union afs_xdr_dir_block),
473 return afs_bad(dvnode, afs_file_error_dir_over_end);
476 /* Check that the name-extension dirents are all allocated */
477 for (tmp = 1; tmp < nr_slots; tmp++) {
478 unsigned int ix = offset + tmp;
479 if (!(block->hdr.bitmap[ix / 8] & (1 << (ix % 8)))) {
481 " %u unmarked extension (%u/%u)",
482 blkoff / sizeof(union afs_xdr_dir_block),
483 offset, tmp, nr_slots);
484 return afs_bad(dvnode, afs_file_error_dir_unmarked_ext);
488 /* skip if starts before the current position */
492 /* found the next entry */
493 if (!dir_emit(ctx, dire->u.name, nlen,
494 ntohl(dire->u.vnode),
495 (ctx->actor == afs_lookup_filldir ||
496 ctx->actor == afs_lookup_one_filldir)?
497 ntohl(dire->u.unique) : DT_UNKNOWN)) {
498 _leave(" = 0 [full]");
502 ctx->pos = blkoff + next * sizeof(union afs_xdr_dirent);
505 _leave(" = 1 [more]");
510 * iterate through the data blob that lists the contents of an AFS directory
512 static int afs_dir_iterate(struct inode *dir, struct dir_context *ctx,
513 struct key *key, afs_dataversion_t *_dir_version)
515 struct afs_vnode *dvnode = AFS_FS_I(dir);
516 struct afs_xdr_dir_page *dbuf;
517 union afs_xdr_dir_block *dblock;
518 struct afs_read *req;
520 unsigned blkoff, limit;
524 _enter("{%lu},%u,,", dir->i_ino, (unsigned)ctx->pos);
526 if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(dir)->flags)) {
527 _leave(" = -ESTALE");
531 req = afs_read_dir(dvnode, key);
534 *_dir_version = req->data_version;
536 /* round the file position up to the next entry boundary */
537 ctx->pos += sizeof(union afs_xdr_dirent) - 1;
538 ctx->pos &= ~(sizeof(union afs_xdr_dirent) - 1);
540 /* walk through the blocks in sequence */
542 while (ctx->pos < req->actual_len) {
543 blkoff = ctx->pos & ~(sizeof(union afs_xdr_dir_block) - 1);
545 /* Fetch the appropriate page from the directory and re-add it
546 * to the LRU. We have all the pages pinned with an extra ref.
550 slot = radix_tree_lookup_slot(&dvnode->vfs_inode.i_mapping->i_pages,
553 page = radix_tree_deref_slot(slot);
556 ret = afs_bad(dvnode, afs_file_error_dir_missing_page);
559 mark_page_accessed(page);
561 limit = blkoff & ~(PAGE_SIZE - 1);
565 /* deal with the individual blocks stashed on this page */
567 dblock = &dbuf->blocks[(blkoff % PAGE_SIZE) /
568 sizeof(union afs_xdr_dir_block)];
569 ret = afs_dir_iterate_block(dvnode, ctx, dblock, blkoff);
575 blkoff += sizeof(union afs_xdr_dir_block);
577 } while (ctx->pos < dir->i_size && blkoff < limit);
584 up_read(&dvnode->validate_lock);
586 _leave(" = %d", ret);
591 * read an AFS directory
593 static int afs_readdir(struct file *file, struct dir_context *ctx)
595 afs_dataversion_t dir_version;
597 return afs_dir_iterate(file_inode(file), ctx, afs_file_key(file),
602 * Search the directory for a single name
603 * - if afs_dir_iterate_block() spots this function, it'll pass the FID
604 * uniquifier through dtype
606 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name,
607 int nlen, loff_t fpos, u64 ino, unsigned dtype)
609 struct afs_lookup_one_cookie *cookie =
610 container_of(ctx, struct afs_lookup_one_cookie, ctx);
612 _enter("{%s,%u},%s,%u,,%llu,%u",
613 cookie->name.name, cookie->name.len, name, nlen,
614 (unsigned long long) ino, dtype);
616 /* insanity checks first */
617 BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
618 BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
620 if (cookie->name.len != nlen ||
621 memcmp(cookie->name.name, name, nlen) != 0) {
626 cookie->fid.vnode = ino;
627 cookie->fid.unique = dtype;
630 _leave(" = -1 [found]");
635 * Do a lookup of a single name in a directory
636 * - just returns the FID the dentry name maps to if found
638 static int afs_do_lookup_one(struct inode *dir, struct dentry *dentry,
639 struct afs_fid *fid, struct key *key,
640 afs_dataversion_t *_dir_version)
642 struct afs_super_info *as = dir->i_sb->s_fs_info;
643 struct afs_lookup_one_cookie cookie = {
644 .ctx.actor = afs_lookup_one_filldir,
645 .name = dentry->d_name,
646 .fid.vid = as->volume->vid
650 _enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
652 /* search the directory */
653 ret = afs_dir_iterate(dir, &cookie.ctx, key, _dir_version);
655 _leave(" = %d [iter]", ret);
661 _leave(" = -ENOENT [not found]");
666 _leave(" = 0 { vn=%llu u=%u }", fid->vnode, fid->unique);
671 * search the directory for a name
672 * - if afs_dir_iterate_block() spots this function, it'll pass the FID
673 * uniquifier through dtype
675 static int afs_lookup_filldir(struct dir_context *ctx, const char *name,
676 int nlen, loff_t fpos, u64 ino, unsigned dtype)
678 struct afs_lookup_cookie *cookie =
679 container_of(ctx, struct afs_lookup_cookie, ctx);
682 _enter("{%s,%u},%s,%u,,%llu,%u",
683 cookie->name.name, cookie->name.len, name, nlen,
684 (unsigned long long) ino, dtype);
686 /* insanity checks first */
687 BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
688 BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
691 if (cookie->nr_fids < 50) {
692 cookie->fids[cookie->nr_fids].vnode = ino;
693 cookie->fids[cookie->nr_fids].unique = dtype;
696 } else if (cookie->name.len == nlen &&
697 memcmp(cookie->name.name, name, nlen) == 0) {
698 cookie->fids[1].vnode = ino;
699 cookie->fids[1].unique = dtype;
701 if (cookie->one_only)
705 ret = cookie->nr_fids >= 50 ? -1 : 0;
706 _leave(" = %d", ret);
711 * Deal with the result of a successful lookup operation. Turn all the files
712 * into inodes and save the first one - which is the one we actually want.
714 static void afs_do_lookup_success(struct afs_operation *op)
716 struct afs_vnode_param *vp;
717 struct afs_vnode *vnode;
724 for (i = 0; i < op->nr_files; i++) {
728 abort_code = vp->scb.status.abort_code;
729 if (abort_code != 0) {
730 op->ac.abort_code = abort_code;
731 op->error = afs_abort_to_error(abort_code);
740 vp = &op->more_files[i - 2];
744 if (!vp->scb.have_status && !vp->scb.have_error)
747 _debug("do [%u]", i);
749 if (!test_bit(AFS_VNODE_UNSET, &vp->vnode->flags))
750 afs_vnode_commit_status(op, vp);
751 } else if (vp->scb.status.abort_code == 0) {
752 inode = afs_iget(op, vp);
753 if (!IS_ERR(inode)) {
754 vnode = AFS_FS_I(inode);
755 afs_cache_permit(vnode, op->key,
756 0 /* Assume vnode->cb_break is 0 */ +
760 vp->put_vnode = true;
763 _debug("- abort %d %llx:%llx.%x",
764 vp->scb.status.abort_code,
765 vp->fid.vid, vp->fid.vnode, vp->fid.unique);
772 static const struct afs_operation_ops afs_inline_bulk_status_operation = {
773 .issue_afs_rpc = afs_fs_inline_bulk_status,
774 .issue_yfs_rpc = yfs_fs_inline_bulk_status,
775 .success = afs_do_lookup_success,
778 static const struct afs_operation_ops afs_lookup_fetch_status_operation = {
779 .issue_afs_rpc = afs_fs_fetch_status,
780 .issue_yfs_rpc = yfs_fs_fetch_status,
781 .success = afs_do_lookup_success,
782 .aborted = afs_check_for_remote_deletion,
786 * See if we know that the server we expect to use doesn't support
787 * FS.InlineBulkStatus.
789 static bool afs_server_supports_ibulk(struct afs_vnode *dvnode)
791 struct afs_server_list *slist;
792 struct afs_volume *volume = dvnode->volume;
793 struct afs_server *server;
797 if (!test_bit(AFS_VOLUME_MAYBE_NO_IBULK, &volume->flags))
801 slist = rcu_dereference(volume->servers);
803 for (i = 0; i < slist->nr_servers; i++) {
804 server = slist->servers[i].server;
805 if (server == dvnode->cb_server) {
806 if (test_bit(AFS_SERVER_FL_NO_IBULK, &server->flags))
817 * Do a lookup in a directory. We make use of bulk lookup to query a slew of
818 * files in one go and create inodes for them. The inode of the file we were
819 * asked for is returned.
821 static struct inode *afs_do_lookup(struct inode *dir, struct dentry *dentry,
824 struct afs_lookup_cookie *cookie;
825 struct afs_vnode_param *vp;
826 struct afs_operation *op;
827 struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode;
828 struct inode *inode = NULL, *ti;
829 afs_dataversion_t data_version = READ_ONCE(dvnode->status.data_version);
833 _enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
835 cookie = kzalloc(sizeof(struct afs_lookup_cookie), GFP_KERNEL);
837 return ERR_PTR(-ENOMEM);
839 for (i = 0; i < ARRAY_SIZE(cookie->fids); i++)
840 cookie->fids[i].vid = dvnode->fid.vid;
841 cookie->ctx.actor = afs_lookup_filldir;
842 cookie->name = dentry->d_name;
843 cookie->nr_fids = 2; /* slot 0 is saved for the fid we actually want
844 * and slot 1 for the directory */
846 if (!afs_server_supports_ibulk(dvnode))
847 cookie->one_only = true;
849 /* search the directory */
850 ret = afs_dir_iterate(dir, &cookie->ctx, key, &data_version);
854 dentry->d_fsdata = (void *)(unsigned long)data_version;
860 /* Check to see if we already have an inode for the primary fid. */
861 inode = ilookup5(dir->i_sb, cookie->fids[1].vnode,
862 afs_ilookup5_test_by_fid, &cookie->fids[1]);
864 goto out; /* We do */
866 /* Okay, we didn't find it. We need to query the server - and whilst
867 * we're doing that, we're going to attempt to look up a bunch of other
870 op = afs_alloc_operation(NULL, dvnode->volume);
876 afs_op_set_vnode(op, 0, dvnode);
877 afs_op_set_fid(op, 1, &cookie->fids[1]);
879 op->nr_files = cookie->nr_fids;
880 _debug("nr_files %u", op->nr_files);
882 /* Need space for examining all the selected files */
884 if (op->nr_files > 2) {
885 op->more_files = kvcalloc(op->nr_files - 2,
886 sizeof(struct afs_vnode_param),
891 for (i = 2; i < op->nr_files; i++) {
892 vp = &op->more_files[i - 2];
893 vp->fid = cookie->fids[i];
895 /* Find any inodes that already exist and get their
898 ti = ilookup5_nowait(dir->i_sb, vp->fid.vnode,
899 afs_ilookup5_test_by_fid, &vp->fid);
900 if (!IS_ERR_OR_NULL(ti)) {
901 vnode = AFS_FS_I(ti);
902 vp->dv_before = vnode->status.data_version;
903 vp->cb_break_before = afs_calc_vnode_cb_break(vnode);
905 vp->put_vnode = true;
906 vp->speculative = true; /* vnode not locked */
911 /* Try FS.InlineBulkStatus first. Abort codes for the individual
912 * lookups contained therein are stored in the reply without aborting
913 * the whole operation.
915 op->error = -ENOTSUPP;
916 if (!cookie->one_only) {
917 op->ops = &afs_inline_bulk_status_operation;
918 afs_begin_vnode_operation(op);
919 afs_wait_for_operation(op);
922 if (op->error == -ENOTSUPP) {
923 /* We could try FS.BulkStatus next, but this aborts the entire
924 * op if any of the lookups fails - so, for the moment, revert
925 * to FS.FetchStatus for op->file[1].
927 op->fetch_status.which = 1;
928 op->ops = &afs_lookup_fetch_status_operation;
929 afs_begin_vnode_operation(op);
930 afs_wait_for_operation(op);
932 inode = ERR_PTR(op->error);
935 if (op->error == 0) {
936 inode = &op->file[1].vnode->vfs_inode;
937 op->file[1].vnode = NULL;
940 if (op->file[0].scb.have_status)
941 dentry->d_fsdata = (void *)(unsigned long)op->file[0].scb.status.data_version;
943 dentry->d_fsdata = (void *)(unsigned long)op->file[0].dv_before;
944 ret = afs_put_operation(op);
948 return inode ?: ERR_PTR(ret);
952 * Look up an entry in a directory with @sys substitution.
954 static struct dentry *afs_lookup_atsys(struct inode *dir, struct dentry *dentry,
957 struct afs_sysnames *subs;
958 struct afs_net *net = afs_i2net(dir);
960 char *buf, *p, *name;
965 ret = ERR_PTR(-ENOMEM);
966 p = buf = kmalloc(AFSNAMEMAX, GFP_KERNEL);
969 if (dentry->d_name.len > 4) {
970 memcpy(p, dentry->d_name.name, dentry->d_name.len - 4);
971 p += dentry->d_name.len - 4;
974 /* There is an ordered list of substitutes that we have to try. */
975 read_lock(&net->sysnames_lock);
976 subs = net->sysnames;
977 refcount_inc(&subs->usage);
978 read_unlock(&net->sysnames_lock);
980 for (i = 0; i < subs->nr; i++) {
981 name = subs->subs[i];
982 len = dentry->d_name.len - 4 + strlen(name);
983 if (len >= AFSNAMEMAX) {
984 ret = ERR_PTR(-ENAMETOOLONG);
989 ret = lookup_one_len(buf, dentry->d_parent, len);
990 if (IS_ERR(ret) || d_is_positive(ret))
995 /* We don't want to d_add() the @sys dentry here as we don't want to
996 * the cached dentry to hide changes to the sysnames list.
1000 afs_put_sysnames(subs);
1008 * look up an entry in a directory
1010 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
1013 struct afs_vnode *dvnode = AFS_FS_I(dir);
1014 struct afs_fid fid = {};
1015 struct inode *inode;
1020 _enter("{%llx:%llu},%p{%pd},",
1021 dvnode->fid.vid, dvnode->fid.vnode, dentry, dentry);
1023 ASSERTCMP(d_inode(dentry), ==, NULL);
1025 if (dentry->d_name.len >= AFSNAMEMAX) {
1026 _leave(" = -ENAMETOOLONG");
1027 return ERR_PTR(-ENAMETOOLONG);
1030 if (test_bit(AFS_VNODE_DELETED, &dvnode->flags)) {
1031 _leave(" = -ESTALE");
1032 return ERR_PTR(-ESTALE);
1035 key = afs_request_key(dvnode->volume->cell);
1037 _leave(" = %ld [key]", PTR_ERR(key));
1038 return ERR_CAST(key);
1041 ret = afs_validate(dvnode, key);
1044 _leave(" = %d [val]", ret);
1045 return ERR_PTR(ret);
1048 if (dentry->d_name.len >= 4 &&
1049 dentry->d_name.name[dentry->d_name.len - 4] == '@' &&
1050 dentry->d_name.name[dentry->d_name.len - 3] == 's' &&
1051 dentry->d_name.name[dentry->d_name.len - 2] == 'y' &&
1052 dentry->d_name.name[dentry->d_name.len - 1] == 's')
1053 return afs_lookup_atsys(dir, dentry, key);
1055 afs_stat_v(dvnode, n_lookup);
1056 inode = afs_do_lookup(dir, dentry, key);
1058 if (inode == ERR_PTR(-ENOENT))
1059 inode = afs_try_auto_mntpt(dentry, dir);
1061 if (!IS_ERR_OR_NULL(inode))
1062 fid = AFS_FS_I(inode)->fid;
1064 _debug("splice %p", dentry->d_inode);
1065 d = d_splice_alias(inode, dentry);
1066 if (!IS_ERR_OR_NULL(d)) {
1067 d->d_fsdata = dentry->d_fsdata;
1068 trace_afs_lookup(dvnode, &d->d_name, &fid);
1070 trace_afs_lookup(dvnode, &dentry->d_name, &fid);
1077 * Check the validity of a dentry under RCU conditions.
1079 static int afs_d_revalidate_rcu(struct dentry *dentry)
1081 struct afs_vnode *dvnode, *vnode;
1082 struct dentry *parent;
1083 struct inode *dir, *inode;
1084 long dir_version, de_version;
1086 _enter("%p", dentry);
1088 /* Check the parent directory is still valid first. */
1089 parent = READ_ONCE(dentry->d_parent);
1090 dir = d_inode_rcu(parent);
1093 dvnode = AFS_FS_I(dir);
1094 if (test_bit(AFS_VNODE_DELETED, &dvnode->flags))
1097 if (!afs_check_validity(dvnode))
1100 /* We only need to invalidate a dentry if the server's copy changed
1101 * behind our back. If we made the change, it's no problem. Note that
1102 * on a 32-bit system, we only have 32 bits in the dentry to store the
1105 dir_version = (long)READ_ONCE(dvnode->status.data_version);
1106 de_version = (long)READ_ONCE(dentry->d_fsdata);
1107 if (de_version != dir_version) {
1108 dir_version = (long)READ_ONCE(dvnode->invalid_before);
1109 if (de_version - dir_version < 0)
1113 /* Check to see if the vnode referred to by the dentry still
1116 if (d_really_is_positive(dentry)) {
1117 inode = d_inode_rcu(dentry);
1119 vnode = AFS_FS_I(inode);
1120 if (!afs_check_validity(vnode))
1125 return 1; /* Still valid */
1129 * check that a dentry lookup hit has found a valid entry
1130 * - NOTE! the hit can be a negative hit too, so we can't assume we have an
1133 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags)
1135 struct afs_vnode *vnode, *dir;
1137 struct dentry *parent;
1138 struct inode *inode;
1140 afs_dataversion_t dir_version, invalid_before;
1144 if (flags & LOOKUP_RCU)
1145 return afs_d_revalidate_rcu(dentry);
1147 if (d_really_is_positive(dentry)) {
1148 vnode = AFS_FS_I(d_inode(dentry));
1149 _enter("{v={%llx:%llu} n=%pd fl=%lx},",
1150 vnode->fid.vid, vnode->fid.vnode, dentry,
1153 _enter("{neg n=%pd}", dentry);
1156 key = afs_request_key(AFS_FS_S(dentry->d_sb)->volume->cell);
1160 if (d_really_is_positive(dentry)) {
1161 inode = d_inode(dentry);
1163 vnode = AFS_FS_I(inode);
1164 afs_validate(vnode, key);
1165 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1170 /* lock down the parent dentry so we can peer at it */
1171 parent = dget_parent(dentry);
1172 dir = AFS_FS_I(d_inode(parent));
1174 /* validate the parent directory */
1175 afs_validate(dir, key);
1177 if (test_bit(AFS_VNODE_DELETED, &dir->flags)) {
1178 _debug("%pd: parent dir deleted", dentry);
1179 goto out_bad_parent;
1182 /* We only need to invalidate a dentry if the server's copy changed
1183 * behind our back. If we made the change, it's no problem. Note that
1184 * on a 32-bit system, we only have 32 bits in the dentry to store the
1187 dir_version = dir->status.data_version;
1188 de_version = (long)dentry->d_fsdata;
1189 if (de_version == (long)dir_version)
1190 goto out_valid_noupdate;
1192 invalid_before = dir->invalid_before;
1193 if (de_version - (long)invalid_before >= 0)
1196 _debug("dir modified");
1197 afs_stat_v(dir, n_reval);
1199 /* search the directory for this vnode */
1200 ret = afs_do_lookup_one(&dir->vfs_inode, dentry, &fid, key, &dir_version);
1203 /* the filename maps to something */
1204 if (d_really_is_negative(dentry))
1205 goto out_bad_parent;
1206 inode = d_inode(dentry);
1207 if (is_bad_inode(inode)) {
1208 printk("kAFS: afs_d_revalidate: %pd2 has bad inode\n",
1210 goto out_bad_parent;
1213 vnode = AFS_FS_I(inode);
1215 /* if the vnode ID has changed, then the dirent points to a
1217 if (fid.vnode != vnode->fid.vnode) {
1218 _debug("%pd: dirent changed [%llu != %llu]",
1224 /* if the vnode ID uniqifier has changed, then the file has
1225 * been deleted and replaced, and the original vnode ID has
1227 if (fid.unique != vnode->fid.unique) {
1228 _debug("%pd: file deleted (uq %u -> %u I:%u)",
1231 vnode->vfs_inode.i_generation);
1232 write_seqlock(&vnode->cb_lock);
1233 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1234 write_sequnlock(&vnode->cb_lock);
1240 /* the filename is unknown */
1241 _debug("%pd: dirent not found", dentry);
1242 if (d_really_is_positive(dentry))
1247 _debug("failed to iterate dir %pd: %d",
1249 goto out_bad_parent;
1253 dentry->d_fsdata = (void *)(unsigned long)dir_version;
1257 _leave(" = 1 [valid]");
1260 /* the dirent, if it exists, now points to a different vnode */
1262 spin_lock(&dentry->d_lock);
1263 dentry->d_flags |= DCACHE_NFSFS_RENAMED;
1264 spin_unlock(&dentry->d_lock);
1267 _debug("dropping dentry %pd2", dentry);
1272 _leave(" = 0 [bad]");
1277 * allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
1279 * - called from dput() when d_count is going to 0.
1280 * - return 1 to request dentry be unhashed, 0 otherwise
1282 static int afs_d_delete(const struct dentry *dentry)
1284 _enter("%pd", dentry);
1286 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1289 if (d_really_is_positive(dentry) &&
1290 (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(d_inode(dentry))->flags) ||
1291 test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(d_inode(dentry))->flags)))
1294 _leave(" = 0 [keep]");
1298 _leave(" = 1 [zap]");
1303 * Clean up sillyrename files on dentry removal.
1305 static void afs_d_iput(struct dentry *dentry, struct inode *inode)
1307 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1308 afs_silly_iput(dentry, inode);
1313 * handle dentry release
1315 void afs_d_release(struct dentry *dentry)
1317 _enter("%pd", dentry);
1320 void afs_check_for_remote_deletion(struct afs_operation *op)
1322 struct afs_vnode *vnode = op->file[0].vnode;
1324 switch (op->ac.abort_code) {
1326 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1327 afs_break_callback(vnode, afs_cb_break_for_deleted);
1332 * Create a new inode for create/mkdir/symlink
1334 static void afs_vnode_new_inode(struct afs_operation *op)
1336 struct afs_vnode_param *vp = &op->file[1];
1337 struct afs_vnode *vnode;
1338 struct inode *inode;
1342 ASSERTCMP(op->error, ==, 0);
1344 inode = afs_iget(op, vp);
1345 if (IS_ERR(inode)) {
1346 /* ENOMEM or EINTR at a really inconvenient time - just abandon
1347 * the new directory on the server.
1349 op->error = PTR_ERR(inode);
1353 vnode = AFS_FS_I(inode);
1354 set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
1356 afs_cache_permit(vnode, op->key, vnode->cb_break, &vp->scb);
1357 d_instantiate(op->dentry, inode);
1360 static void afs_create_success(struct afs_operation *op)
1362 _enter("op=%08x", op->debug_id);
1363 op->ctime = op->file[0].scb.status.mtime_client;
1364 afs_vnode_commit_status(op, &op->file[0]);
1365 afs_update_dentry_version(op, &op->file[0], op->dentry);
1366 afs_vnode_new_inode(op);
1369 static void afs_create_edit_dir(struct afs_operation *op)
1371 struct afs_vnode_param *dvp = &op->file[0];
1372 struct afs_vnode_param *vp = &op->file[1];
1373 struct afs_vnode *dvnode = dvp->vnode;
1375 _enter("op=%08x", op->debug_id);
1377 down_write(&dvnode->validate_lock);
1378 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1379 dvnode->status.data_version == dvp->dv_before + dvp->dv_delta)
1380 afs_edit_dir_add(dvnode, &op->dentry->d_name, &vp->fid,
1382 up_write(&dvnode->validate_lock);
1385 static void afs_create_put(struct afs_operation *op)
1387 _enter("op=%08x", op->debug_id);
1393 static const struct afs_operation_ops afs_mkdir_operation = {
1394 .issue_afs_rpc = afs_fs_make_dir,
1395 .issue_yfs_rpc = yfs_fs_make_dir,
1396 .success = afs_create_success,
1397 .aborted = afs_check_for_remote_deletion,
1398 .edit_dir = afs_create_edit_dir,
1399 .put = afs_create_put,
1403 * create a directory on an AFS filesystem
1405 static int afs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1406 struct dentry *dentry, umode_t mode)
1408 struct afs_operation *op;
1409 struct afs_vnode *dvnode = AFS_FS_I(dir);
1411 _enter("{%llx:%llu},{%pd},%ho",
1412 dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1414 op = afs_alloc_operation(NULL, dvnode->volume);
1420 afs_op_set_vnode(op, 0, dvnode);
1421 op->file[0].dv_delta = 1;
1422 op->file[0].modification = true;
1423 op->file[0].update_ctime = true;
1424 op->dentry = dentry;
1425 op->create.mode = S_IFDIR | mode;
1426 op->create.reason = afs_edit_dir_for_mkdir;
1427 op->ops = &afs_mkdir_operation;
1428 return afs_do_sync_operation(op);
1432 * Remove a subdir from a directory.
1434 static void afs_dir_remove_subdir(struct dentry *dentry)
1436 if (d_really_is_positive(dentry)) {
1437 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1439 clear_nlink(&vnode->vfs_inode);
1440 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1441 clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1442 clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
1446 static void afs_rmdir_success(struct afs_operation *op)
1448 _enter("op=%08x", op->debug_id);
1449 op->ctime = op->file[0].scb.status.mtime_client;
1450 afs_vnode_commit_status(op, &op->file[0]);
1451 afs_update_dentry_version(op, &op->file[0], op->dentry);
1454 static void afs_rmdir_edit_dir(struct afs_operation *op)
1456 struct afs_vnode_param *dvp = &op->file[0];
1457 struct afs_vnode *dvnode = dvp->vnode;
1459 _enter("op=%08x", op->debug_id);
1460 afs_dir_remove_subdir(op->dentry);
1462 down_write(&dvnode->validate_lock);
1463 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1464 dvnode->status.data_version == dvp->dv_before + dvp->dv_delta)
1465 afs_edit_dir_remove(dvnode, &op->dentry->d_name,
1466 afs_edit_dir_for_rmdir);
1467 up_write(&dvnode->validate_lock);
1470 static void afs_rmdir_put(struct afs_operation *op)
1472 _enter("op=%08x", op->debug_id);
1473 if (op->file[1].vnode)
1474 up_write(&op->file[1].vnode->rmdir_lock);
1477 static const struct afs_operation_ops afs_rmdir_operation = {
1478 .issue_afs_rpc = afs_fs_remove_dir,
1479 .issue_yfs_rpc = yfs_fs_remove_dir,
1480 .success = afs_rmdir_success,
1481 .aborted = afs_check_for_remote_deletion,
1482 .edit_dir = afs_rmdir_edit_dir,
1483 .put = afs_rmdir_put,
1487 * remove a directory from an AFS filesystem
1489 static int afs_rmdir(struct inode *dir, struct dentry *dentry)
1491 struct afs_operation *op;
1492 struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode = NULL;
1495 _enter("{%llx:%llu},{%pd}",
1496 dvnode->fid.vid, dvnode->fid.vnode, dentry);
1498 op = afs_alloc_operation(NULL, dvnode->volume);
1502 afs_op_set_vnode(op, 0, dvnode);
1503 op->file[0].dv_delta = 1;
1504 op->file[0].modification = true;
1505 op->file[0].update_ctime = true;
1507 op->dentry = dentry;
1508 op->ops = &afs_rmdir_operation;
1510 /* Try to make sure we have a callback promise on the victim. */
1511 if (d_really_is_positive(dentry)) {
1512 vnode = AFS_FS_I(d_inode(dentry));
1513 ret = afs_validate(vnode, op->key);
1519 ret = down_write_killable(&vnode->rmdir_lock);
1522 op->file[1].vnode = vnode;
1525 return afs_do_sync_operation(op);
1528 return afs_put_operation(op);
1532 * Remove a link to a file or symlink from a directory.
1534 * If the file was not deleted due to excess hard links, the fileserver will
1535 * break the callback promise on the file - if it had one - before it returns
1536 * to us, and if it was deleted, it won't
1538 * However, if we didn't have a callback promise outstanding, or it was
1539 * outstanding on a different server, then it won't break it either...
1541 static void afs_dir_remove_link(struct afs_operation *op)
1543 struct afs_vnode *dvnode = op->file[0].vnode;
1544 struct afs_vnode *vnode = op->file[1].vnode;
1545 struct dentry *dentry = op->dentry;
1548 if (op->error != 0 ||
1549 (op->file[1].scb.have_status && op->file[1].scb.have_error))
1551 if (d_really_is_positive(dentry))
1554 if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
1556 } else if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
1557 write_seqlock(&vnode->cb_lock);
1558 drop_nlink(&vnode->vfs_inode);
1559 if (vnode->vfs_inode.i_nlink == 0) {
1560 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1561 __afs_break_callback(vnode, afs_cb_break_for_unlink);
1563 write_sequnlock(&vnode->cb_lock);
1565 afs_break_callback(vnode, afs_cb_break_for_unlink);
1567 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1568 _debug("AFS_VNODE_DELETED");
1570 ret = afs_validate(vnode, op->key);
1575 _debug("nlink %d [val %d]", vnode->vfs_inode.i_nlink, op->error);
1578 static void afs_unlink_success(struct afs_operation *op)
1580 _enter("op=%08x", op->debug_id);
1581 op->ctime = op->file[0].scb.status.mtime_client;
1582 afs_check_dir_conflict(op, &op->file[0]);
1583 afs_vnode_commit_status(op, &op->file[0]);
1584 afs_vnode_commit_status(op, &op->file[1]);
1585 afs_update_dentry_version(op, &op->file[0], op->dentry);
1586 afs_dir_remove_link(op);
1589 static void afs_unlink_edit_dir(struct afs_operation *op)
1591 struct afs_vnode_param *dvp = &op->file[0];
1592 struct afs_vnode *dvnode = dvp->vnode;
1594 _enter("op=%08x", op->debug_id);
1595 down_write(&dvnode->validate_lock);
1596 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags) &&
1597 dvnode->status.data_version == dvp->dv_before + dvp->dv_delta)
1598 afs_edit_dir_remove(dvnode, &op->dentry->d_name,
1599 afs_edit_dir_for_unlink);
1600 up_write(&dvnode->validate_lock);
1603 static void afs_unlink_put(struct afs_operation *op)
1605 _enter("op=%08x", op->debug_id);
1606 if (op->unlink.need_rehash && op->error < 0 && op->error != -ENOENT)
1607 d_rehash(op->dentry);
1610 static const struct afs_operation_ops afs_unlink_operation = {
1611 .issue_afs_rpc = afs_fs_remove_file,
1612 .issue_yfs_rpc = yfs_fs_remove_file,
1613 .success = afs_unlink_success,
1614 .aborted = afs_check_for_remote_deletion,
1615 .edit_dir = afs_unlink_edit_dir,
1616 .put = afs_unlink_put,
1620 * Remove a file or symlink from an AFS filesystem.
1622 static int afs_unlink(struct inode *dir, struct dentry *dentry)
1624 struct afs_operation *op;
1625 struct afs_vnode *dvnode = AFS_FS_I(dir);
1626 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1629 _enter("{%llx:%llu},{%pd}",
1630 dvnode->fid.vid, dvnode->fid.vnode, dentry);
1632 if (dentry->d_name.len >= AFSNAMEMAX)
1633 return -ENAMETOOLONG;
1635 op = afs_alloc_operation(NULL, dvnode->volume);
1639 afs_op_set_vnode(op, 0, dvnode);
1640 op->file[0].dv_delta = 1;
1641 op->file[0].modification = true;
1642 op->file[0].update_ctime = true;
1644 /* Try to make sure we have a callback promise on the victim. */
1645 ret = afs_validate(vnode, op->key);
1651 spin_lock(&dentry->d_lock);
1652 if (d_count(dentry) > 1) {
1653 spin_unlock(&dentry->d_lock);
1654 /* Start asynchronous writeout of the inode */
1655 write_inode_now(d_inode(dentry), 0);
1656 op->error = afs_sillyrename(dvnode, vnode, dentry, op->key);
1659 if (!d_unhashed(dentry)) {
1660 /* Prevent a race with RCU lookup. */
1662 op->unlink.need_rehash = true;
1664 spin_unlock(&dentry->d_lock);
1666 op->file[1].vnode = vnode;
1667 op->file[1].update_ctime = true;
1668 op->file[1].op_unlinked = true;
1669 op->dentry = dentry;
1670 op->ops = &afs_unlink_operation;
1671 afs_begin_vnode_operation(op);
1672 afs_wait_for_operation(op);
1674 /* If there was a conflict with a third party, check the status of the
1677 if (op->error == 0 && (op->flags & AFS_OPERATION_DIR_CONFLICT)) {
1678 op->file[1].update_ctime = false;
1679 op->fetch_status.which = 1;
1680 op->ops = &afs_fetch_status_operation;
1681 afs_begin_vnode_operation(op);
1682 afs_wait_for_operation(op);
1685 return afs_put_operation(op);
1688 return afs_put_operation(op);
1691 static const struct afs_operation_ops afs_create_operation = {
1692 .issue_afs_rpc = afs_fs_create_file,
1693 .issue_yfs_rpc = yfs_fs_create_file,
1694 .success = afs_create_success,
1695 .aborted = afs_check_for_remote_deletion,
1696 .edit_dir = afs_create_edit_dir,
1697 .put = afs_create_put,
1701 * create a regular file on an AFS filesystem
1703 static int afs_create(struct user_namespace *mnt_userns, struct inode *dir,
1704 struct dentry *dentry, umode_t mode, bool excl)
1706 struct afs_operation *op;
1707 struct afs_vnode *dvnode = AFS_FS_I(dir);
1708 int ret = -ENAMETOOLONG;
1710 _enter("{%llx:%llu},{%pd},%ho",
1711 dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1713 if (dentry->d_name.len >= AFSNAMEMAX)
1716 op = afs_alloc_operation(NULL, dvnode->volume);
1722 afs_op_set_vnode(op, 0, dvnode);
1723 op->file[0].dv_delta = 1;
1724 op->file[0].modification = true;
1725 op->file[0].update_ctime = true;
1727 op->dentry = dentry;
1728 op->create.mode = S_IFREG | mode;
1729 op->create.reason = afs_edit_dir_for_create;
1730 op->ops = &afs_create_operation;
1731 return afs_do_sync_operation(op);
1735 _leave(" = %d", ret);
1739 static void afs_link_success(struct afs_operation *op)
1741 struct afs_vnode_param *dvp = &op->file[0];
1742 struct afs_vnode_param *vp = &op->file[1];
1744 _enter("op=%08x", op->debug_id);
1745 op->ctime = dvp->scb.status.mtime_client;
1746 afs_vnode_commit_status(op, dvp);
1747 afs_vnode_commit_status(op, vp);
1748 afs_update_dentry_version(op, dvp, op->dentry);
1749 if (op->dentry_2->d_parent == op->dentry->d_parent)
1750 afs_update_dentry_version(op, dvp, op->dentry_2);
1751 ihold(&vp->vnode->vfs_inode);
1752 d_instantiate(op->dentry, &vp->vnode->vfs_inode);
1755 static void afs_link_put(struct afs_operation *op)
1757 _enter("op=%08x", op->debug_id);
1762 static const struct afs_operation_ops afs_link_operation = {
1763 .issue_afs_rpc = afs_fs_link,
1764 .issue_yfs_rpc = yfs_fs_link,
1765 .success = afs_link_success,
1766 .aborted = afs_check_for_remote_deletion,
1767 .edit_dir = afs_create_edit_dir,
1768 .put = afs_link_put,
1772 * create a hard link between files in an AFS filesystem
1774 static int afs_link(struct dentry *from, struct inode *dir,
1775 struct dentry *dentry)
1777 struct afs_operation *op;
1778 struct afs_vnode *dvnode = AFS_FS_I(dir);
1779 struct afs_vnode *vnode = AFS_FS_I(d_inode(from));
1780 int ret = -ENAMETOOLONG;
1782 _enter("{%llx:%llu},{%llx:%llu},{%pd}",
1783 vnode->fid.vid, vnode->fid.vnode,
1784 dvnode->fid.vid, dvnode->fid.vnode,
1787 if (dentry->d_name.len >= AFSNAMEMAX)
1790 op = afs_alloc_operation(NULL, dvnode->volume);
1796 afs_op_set_vnode(op, 0, dvnode);
1797 afs_op_set_vnode(op, 1, vnode);
1798 op->file[0].dv_delta = 1;
1799 op->file[0].modification = true;
1800 op->file[0].update_ctime = true;
1801 op->file[1].update_ctime = true;
1803 op->dentry = dentry;
1804 op->dentry_2 = from;
1805 op->ops = &afs_link_operation;
1806 op->create.reason = afs_edit_dir_for_link;
1807 return afs_do_sync_operation(op);
1811 _leave(" = %d", ret);
1815 static const struct afs_operation_ops afs_symlink_operation = {
1816 .issue_afs_rpc = afs_fs_symlink,
1817 .issue_yfs_rpc = yfs_fs_symlink,
1818 .success = afs_create_success,
1819 .aborted = afs_check_for_remote_deletion,
1820 .edit_dir = afs_create_edit_dir,
1821 .put = afs_create_put,
1825 * create a symlink in an AFS filesystem
1827 static int afs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1828 struct dentry *dentry, const char *content)
1830 struct afs_operation *op;
1831 struct afs_vnode *dvnode = AFS_FS_I(dir);
1834 _enter("{%llx:%llu},{%pd},%s",
1835 dvnode->fid.vid, dvnode->fid.vnode, dentry,
1838 ret = -ENAMETOOLONG;
1839 if (dentry->d_name.len >= AFSNAMEMAX)
1843 if (strlen(content) >= AFSPATHMAX)
1846 op = afs_alloc_operation(NULL, dvnode->volume);
1852 afs_op_set_vnode(op, 0, dvnode);
1853 op->file[0].dv_delta = 1;
1855 op->dentry = dentry;
1856 op->ops = &afs_symlink_operation;
1857 op->create.reason = afs_edit_dir_for_symlink;
1858 op->create.symlink = content;
1859 return afs_do_sync_operation(op);
1863 _leave(" = %d", ret);
1867 static void afs_rename_success(struct afs_operation *op)
1869 _enter("op=%08x", op->debug_id);
1871 op->ctime = op->file[0].scb.status.mtime_client;
1872 afs_check_dir_conflict(op, &op->file[1]);
1873 afs_vnode_commit_status(op, &op->file[0]);
1874 if (op->file[1].vnode != op->file[0].vnode) {
1875 op->ctime = op->file[1].scb.status.mtime_client;
1876 afs_vnode_commit_status(op, &op->file[1]);
1880 static void afs_rename_edit_dir(struct afs_operation *op)
1882 struct afs_vnode_param *orig_dvp = &op->file[0];
1883 struct afs_vnode_param *new_dvp = &op->file[1];
1884 struct afs_vnode *orig_dvnode = orig_dvp->vnode;
1885 struct afs_vnode *new_dvnode = new_dvp->vnode;
1886 struct afs_vnode *vnode = AFS_FS_I(d_inode(op->dentry));
1887 struct dentry *old_dentry = op->dentry;
1888 struct dentry *new_dentry = op->dentry_2;
1889 struct inode *new_inode;
1891 _enter("op=%08x", op->debug_id);
1893 if (op->rename.rehash) {
1894 d_rehash(op->rename.rehash);
1895 op->rename.rehash = NULL;
1898 down_write(&orig_dvnode->validate_lock);
1899 if (test_bit(AFS_VNODE_DIR_VALID, &orig_dvnode->flags) &&
1900 orig_dvnode->status.data_version == orig_dvp->dv_before + orig_dvp->dv_delta)
1901 afs_edit_dir_remove(orig_dvnode, &old_dentry->d_name,
1902 afs_edit_dir_for_rename_0);
1904 if (new_dvnode != orig_dvnode) {
1905 up_write(&orig_dvnode->validate_lock);
1906 down_write(&new_dvnode->validate_lock);
1909 if (test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags) &&
1910 new_dvnode->status.data_version == new_dvp->dv_before + new_dvp->dv_delta) {
1911 if (!op->rename.new_negative)
1912 afs_edit_dir_remove(new_dvnode, &new_dentry->d_name,
1913 afs_edit_dir_for_rename_1);
1915 afs_edit_dir_add(new_dvnode, &new_dentry->d_name,
1916 &vnode->fid, afs_edit_dir_for_rename_2);
1919 new_inode = d_inode(new_dentry);
1921 spin_lock(&new_inode->i_lock);
1922 if (S_ISDIR(new_inode->i_mode))
1923 clear_nlink(new_inode);
1924 else if (new_inode->i_nlink > 0)
1925 drop_nlink(new_inode);
1926 spin_unlock(&new_inode->i_lock);
1929 /* Now we can update d_fsdata on the dentries to reflect their
1930 * new parent's data_version.
1932 * Note that if we ever implement RENAME_EXCHANGE, we'll have
1933 * to update both dentries with opposing dir versions.
1935 afs_update_dentry_version(op, new_dvp, op->dentry);
1936 afs_update_dentry_version(op, new_dvp, op->dentry_2);
1938 d_move(old_dentry, new_dentry);
1940 up_write(&new_dvnode->validate_lock);
1943 static void afs_rename_put(struct afs_operation *op)
1945 _enter("op=%08x", op->debug_id);
1946 if (op->rename.rehash)
1947 d_rehash(op->rename.rehash);
1948 dput(op->rename.tmp);
1950 d_rehash(op->dentry);
1953 static const struct afs_operation_ops afs_rename_operation = {
1954 .issue_afs_rpc = afs_fs_rename,
1955 .issue_yfs_rpc = yfs_fs_rename,
1956 .success = afs_rename_success,
1957 .edit_dir = afs_rename_edit_dir,
1958 .put = afs_rename_put,
1962 * rename a file in an AFS filesystem and/or move it between directories
1964 static int afs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
1965 struct dentry *old_dentry, struct inode *new_dir,
1966 struct dentry *new_dentry, unsigned int flags)
1968 struct afs_operation *op;
1969 struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
1975 /* Don't allow silly-rename files be moved around. */
1976 if (old_dentry->d_flags & DCACHE_NFSFS_RENAMED)
1979 vnode = AFS_FS_I(d_inode(old_dentry));
1980 orig_dvnode = AFS_FS_I(old_dir);
1981 new_dvnode = AFS_FS_I(new_dir);
1983 _enter("{%llx:%llu},{%llx:%llu},{%llx:%llu},{%pd}",
1984 orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
1985 vnode->fid.vid, vnode->fid.vnode,
1986 new_dvnode->fid.vid, new_dvnode->fid.vnode,
1989 op = afs_alloc_operation(NULL, orig_dvnode->volume);
1993 afs_op_set_vnode(op, 0, orig_dvnode);
1994 afs_op_set_vnode(op, 1, new_dvnode); /* May be same as orig_dvnode */
1995 op->file[0].dv_delta = 1;
1996 op->file[1].dv_delta = 1;
1997 op->file[0].modification = true;
1998 op->file[1].modification = true;
1999 op->file[0].update_ctime = true;
2000 op->file[1].update_ctime = true;
2002 op->dentry = old_dentry;
2003 op->dentry_2 = new_dentry;
2004 op->rename.new_negative = d_is_negative(new_dentry);
2005 op->ops = &afs_rename_operation;
2007 /* For non-directories, check whether the target is busy and if so,
2008 * make a copy of the dentry and then do a silly-rename. If the
2009 * silly-rename succeeds, the copied dentry is hashed and becomes the
2012 if (d_is_positive(new_dentry) && !d_is_dir(new_dentry)) {
2013 /* To prevent any new references to the target during the
2014 * rename, we unhash the dentry in advance.
2016 if (!d_unhashed(new_dentry)) {
2018 op->rename.rehash = new_dentry;
2021 if (d_count(new_dentry) > 2) {
2022 /* copy the target dentry's name */
2024 op->rename.tmp = d_alloc(new_dentry->d_parent,
2025 &new_dentry->d_name);
2026 if (!op->rename.tmp)
2029 ret = afs_sillyrename(new_dvnode,
2030 AFS_FS_I(d_inode(new_dentry)),
2031 new_dentry, op->key);
2035 op->dentry_2 = op->rename.tmp;
2036 op->rename.rehash = NULL;
2037 op->rename.new_negative = true;
2041 /* This bit is potentially nasty as there's a potential race with
2042 * afs_d_revalidate{,_rcu}(). We have to change d_fsdata on the dentry
2043 * to reflect it's new parent's new data_version after the op, but
2044 * d_revalidate may see old_dentry between the op having taken place
2045 * and the version being updated.
2047 * So drop the old_dentry for now to make other threads go through
2048 * lookup instead - which we hold a lock against.
2052 return afs_do_sync_operation(op);
2055 return afs_put_operation(op);
2059 * Release a directory page and clean up its private state if it's not busy
2060 * - return true if the page can now be released, false if not
2062 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags)
2064 struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
2066 _enter("{{%llx:%llu}[%lu]}", dvnode->fid.vid, dvnode->fid.vnode, page->index);
2068 detach_page_private(page);
2070 /* The directory will need reloading. */
2071 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2072 afs_stat_v(dvnode, n_relpg);
2077 * invalidate part or all of a page
2078 * - release a page and clean up its private data if offset is 0 (indicating
2081 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
2082 unsigned int length)
2084 struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
2086 _enter("{%lu},%u,%u", page->index, offset, length);
2088 BUG_ON(!PageLocked(page));
2090 /* The directory will need reloading. */
2091 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
2092 afs_stat_v(dvnode, n_inval);
2094 /* we clean up only if the entire page is being invalidated */
2095 if (offset == 0 && length == thp_size(page))
2096 detach_page_private(page);