1 /* dir.c: AFS filesystem directory handling
3 * Copyright (C) 2002, 2018 Red Hat, Inc. All Rights Reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/kernel.h>
14 #include <linux/namei.h>
15 #include <linux/pagemap.h>
16 #include <linux/swap.h>
17 #include <linux/ctype.h>
18 #include <linux/sched.h>
19 #include <linux/task_io_accounting_ops.h>
23 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
25 static int afs_dir_open(struct inode *inode, struct file *file);
26 static int afs_readdir(struct file *file, struct dir_context *ctx);
27 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags);
28 static int afs_d_delete(const struct dentry *dentry);
29 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name, int nlen,
30 loff_t fpos, u64 ino, unsigned dtype);
31 static int afs_lookup_filldir(struct dir_context *ctx, const char *name, int nlen,
32 loff_t fpos, u64 ino, unsigned dtype);
33 static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
35 static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
36 static int afs_rmdir(struct inode *dir, struct dentry *dentry);
37 static int afs_unlink(struct inode *dir, struct dentry *dentry);
38 static int afs_link(struct dentry *from, struct inode *dir,
39 struct dentry *dentry);
40 static int afs_symlink(struct inode *dir, struct dentry *dentry,
42 static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
43 struct inode *new_dir, struct dentry *new_dentry,
45 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags);
46 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
49 static int afs_dir_set_page_dirty(struct page *page)
51 BUG(); /* This should never happen. */
54 const struct file_operations afs_dir_file_operations = {
56 .release = afs_release,
57 .iterate_shared = afs_readdir,
59 .llseek = generic_file_llseek,
62 const struct inode_operations afs_dir_inode_operations = {
67 .symlink = afs_symlink,
71 .permission = afs_permission,
72 .getattr = afs_getattr,
73 .setattr = afs_setattr,
74 .listxattr = afs_listxattr,
77 const struct address_space_operations afs_dir_aops = {
78 .set_page_dirty = afs_dir_set_page_dirty,
79 .releasepage = afs_dir_releasepage,
80 .invalidatepage = afs_dir_invalidatepage,
83 const struct dentry_operations afs_fs_dentry_operations = {
84 .d_revalidate = afs_d_revalidate,
85 .d_delete = afs_d_delete,
86 .d_release = afs_d_release,
87 .d_automount = afs_d_automount,
90 struct afs_lookup_one_cookie {
91 struct dir_context ctx;
97 struct afs_lookup_cookie {
98 struct dir_context ctx;
102 unsigned short nr_fids;
103 struct afs_file_status *statuses;
104 struct afs_callback *callbacks;
105 struct afs_fid fids[50];
109 * check that a directory page is valid
111 static bool afs_dir_check_page(struct afs_vnode *dvnode, struct page *page,
114 struct afs_xdr_dir_page *dbuf;
118 /* Determine how many magic numbers there should be in this page, but
119 * we must take care because the directory may change size under us.
121 off = page_offset(page);
125 latter = i_size - off;
126 if (latter >= PAGE_SIZE)
130 qty /= sizeof(union afs_xdr_dir_block);
134 for (tmp = 0; tmp < qty; tmp++) {
135 if (dbuf->blocks[tmp].hdr.magic != AFS_DIR_MAGIC) {
136 printk("kAFS: %s(%lx): bad magic %d/%d is %04hx\n",
137 __func__, dvnode->vfs_inode.i_ino, tmp, qty,
138 ntohs(dbuf->blocks[tmp].hdr.magic));
139 trace_afs_dir_check_failed(dvnode, off, i_size);
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;
154 afs_stat_v(dvnode, n_read_dir);
162 * open an AFS directory file
164 static int afs_dir_open(struct inode *inode, struct file *file)
166 _enter("{%lu}", inode->i_ino);
168 BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
169 BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
171 if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(inode)->flags))
174 return afs_open(inode, file);
178 * Read the directory into the pagecache in one go, scrubbing the previous
179 * contents. The list of pages is returned, pinning them so that they don't
180 * get reclaimed during the iteration.
182 static struct afs_read *afs_read_dir(struct afs_vnode *dvnode, struct key *key)
183 __acquires(&dvnode->validate_lock)
185 struct afs_read *req;
187 int nr_pages, nr_inline, i, n;
191 i_size = i_size_read(&dvnode->vfs_inode);
193 return ERR_PTR(-EIO);
194 if (i_size > 2048 * 1024)
195 return ERR_PTR(-EFBIG);
197 _enter("%llu", i_size);
199 /* Get a request record to hold the page list. We want to hold it
200 * inline if we can, but we don't want to make an order 1 allocation.
202 nr_pages = (i_size + PAGE_SIZE - 1) / PAGE_SIZE;
203 nr_inline = nr_pages;
204 if (nr_inline > (PAGE_SIZE - sizeof(*req)) / sizeof(struct page *))
207 req = kzalloc(sizeof(*req) + sizeof(struct page *) * nr_inline,
210 return ERR_PTR(-ENOMEM);
212 refcount_set(&req->usage, 1);
213 req->nr_pages = nr_pages;
214 req->actual_len = i_size; /* May change */
215 req->len = nr_pages * PAGE_SIZE; /* We can ask for more than there is */
216 req->data_version = dvnode->status.data_version; /* May change */
218 req->pages = req->array;
220 req->pages = kcalloc(nr_pages, sizeof(struct page *),
226 /* Get a list of all the pages that hold or will hold the directory
227 * content. We need to fill in any gaps that we might find where the
228 * memory reclaimer has been at work. If there are any gaps, we will
229 * need to reread the entire directory contents.
233 n = find_get_pages_contig(dvnode->vfs_inode.i_mapping, i,
236 _debug("find %u at %u/%u", n, i, req->nr_pages);
238 gfp_t gfp = dvnode->vfs_inode.i_mapping->gfp_mask;
240 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
241 afs_stat_v(dvnode, n_inval);
244 req->pages[i] = __page_cache_alloc(gfp);
247 ret = add_to_page_cache_lru(req->pages[i],
248 dvnode->vfs_inode.i_mapping,
253 set_page_private(req->pages[i], 1);
254 SetPagePrivate(req->pages[i]);
255 unlock_page(req->pages[i]);
260 } while (i < req->nr_pages);
262 /* If we're going to reload, we need to lock all the pages to prevent
266 if (down_read_killable(&dvnode->validate_lock) < 0)
269 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
272 up_read(&dvnode->validate_lock);
273 if (down_write_killable(&dvnode->validate_lock) < 0)
276 if (!test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags)) {
277 ret = afs_fetch_data(dvnode, key, req);
281 task_io_account_read(PAGE_SIZE * req->nr_pages);
283 if (req->len < req->file_size)
284 goto content_has_grown;
286 /* Validate the data we just read. */
288 for (i = 0; i < req->nr_pages; i++)
289 if (!afs_dir_check_page(dvnode, req->pages[i],
293 // TODO: Trim excess pages
295 set_bit(AFS_VNODE_DIR_VALID, &dvnode->flags);
298 downgrade_write(&dvnode->validate_lock);
303 up_write(&dvnode->validate_lock);
306 _leave(" = %d", ret);
310 up_write(&dvnode->validate_lock);
316 * deal with one block in an AFS directory
318 static int afs_dir_iterate_block(struct dir_context *ctx,
319 union afs_xdr_dir_block *block,
322 union afs_xdr_dirent *dire;
323 unsigned offset, next, curr;
327 _enter("%u,%x,%p,,",(unsigned)ctx->pos,blkoff,block);
329 curr = (ctx->pos - blkoff) / sizeof(union afs_xdr_dirent);
331 /* walk through the block, an entry at a time */
332 for (offset = (blkoff == 0 ? AFS_DIR_RESV_BLOCKS0 : AFS_DIR_RESV_BLOCKS);
333 offset < AFS_DIR_SLOTS_PER_BLOCK;
338 /* skip entries marked unused in the bitmap */
339 if (!(block->hdr.bitmap[offset / 8] &
340 (1 << (offset % 8)))) {
341 _debug("ENT[%zu.%u]: unused",
342 blkoff / sizeof(union afs_xdr_dir_block), offset);
345 next * sizeof(union afs_xdr_dirent);
349 /* got a valid entry */
350 dire = &block->dirents[offset];
351 nlen = strnlen(dire->u.name,
353 offset * sizeof(union afs_xdr_dirent));
355 _debug("ENT[%zu.%u]: %s %zu \"%s\"",
356 blkoff / sizeof(union afs_xdr_dir_block), offset,
357 (offset < curr ? "skip" : "fill"),
360 /* work out where the next possible entry is */
361 for (tmp = nlen; tmp > 15; tmp -= sizeof(union afs_xdr_dirent)) {
362 if (next >= AFS_DIR_SLOTS_PER_BLOCK) {
363 _debug("ENT[%zu.%u]:"
364 " %u travelled beyond end dir block"
366 blkoff / sizeof(union afs_xdr_dir_block),
367 offset, next, tmp, nlen);
370 if (!(block->hdr.bitmap[next / 8] &
371 (1 << (next % 8)))) {
372 _debug("ENT[%zu.%u]:"
373 " %u unmarked extension (len %u/%zu)",
374 blkoff / sizeof(union afs_xdr_dir_block),
375 offset, next, tmp, nlen);
379 _debug("ENT[%zu.%u]: ext %u/%zu",
380 blkoff / sizeof(union afs_xdr_dir_block),
385 /* skip if starts before the current position */
389 /* found the next entry */
390 if (!dir_emit(ctx, dire->u.name, nlen,
391 ntohl(dire->u.vnode),
392 (ctx->actor == afs_lookup_filldir ||
393 ctx->actor == afs_lookup_one_filldir)?
394 ntohl(dire->u.unique) : DT_UNKNOWN)) {
395 _leave(" = 0 [full]");
399 ctx->pos = blkoff + next * sizeof(union afs_xdr_dirent);
402 _leave(" = 1 [more]");
407 * iterate through the data blob that lists the contents of an AFS directory
409 static int afs_dir_iterate(struct inode *dir, struct dir_context *ctx,
412 struct afs_vnode *dvnode = AFS_FS_I(dir);
413 struct afs_xdr_dir_page *dbuf;
414 union afs_xdr_dir_block *dblock;
415 struct afs_read *req;
417 unsigned blkoff, limit;
420 _enter("{%lu},%u,,", dir->i_ino, (unsigned)ctx->pos);
422 if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(dir)->flags)) {
423 _leave(" = -ESTALE");
427 req = afs_read_dir(dvnode, key);
431 /* round the file position up to the next entry boundary */
432 ctx->pos += sizeof(union afs_xdr_dirent) - 1;
433 ctx->pos &= ~(sizeof(union afs_xdr_dirent) - 1);
435 /* walk through the blocks in sequence */
437 while (ctx->pos < req->actual_len) {
438 blkoff = ctx->pos & ~(sizeof(union afs_xdr_dir_block) - 1);
440 /* Fetch the appropriate page from the directory and re-add it
443 page = req->pages[blkoff / PAGE_SIZE];
448 mark_page_accessed(page);
450 limit = blkoff & ~(PAGE_SIZE - 1);
454 /* deal with the individual blocks stashed on this page */
456 dblock = &dbuf->blocks[(blkoff % PAGE_SIZE) /
457 sizeof(union afs_xdr_dir_block)];
458 ret = afs_dir_iterate_block(ctx, dblock, blkoff);
464 blkoff += sizeof(union afs_xdr_dir_block);
466 } while (ctx->pos < dir->i_size && blkoff < limit);
473 up_read(&dvnode->validate_lock);
475 _leave(" = %d", ret);
480 * read an AFS directory
482 static int afs_readdir(struct file *file, struct dir_context *ctx)
484 return afs_dir_iterate(file_inode(file), ctx, afs_file_key(file));
488 * Search the directory for a single name
489 * - if afs_dir_iterate_block() spots this function, it'll pass the FID
490 * uniquifier through dtype
492 static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name,
493 int nlen, loff_t fpos, u64 ino, unsigned dtype)
495 struct afs_lookup_one_cookie *cookie =
496 container_of(ctx, struct afs_lookup_one_cookie, ctx);
498 _enter("{%s,%u},%s,%u,,%llu,%u",
499 cookie->name.name, cookie->name.len, name, nlen,
500 (unsigned long long) ino, dtype);
502 /* insanity checks first */
503 BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
504 BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
506 if (cookie->name.len != nlen ||
507 memcmp(cookie->name.name, name, nlen) != 0) {
512 cookie->fid.vnode = ino;
513 cookie->fid.unique = dtype;
516 _leave(" = -1 [found]");
521 * Do a lookup of a single name in a directory
522 * - just returns the FID the dentry name maps to if found
524 static int afs_do_lookup_one(struct inode *dir, struct dentry *dentry,
525 struct afs_fid *fid, struct key *key)
527 struct afs_super_info *as = dir->i_sb->s_fs_info;
528 struct afs_lookup_one_cookie cookie = {
529 .ctx.actor = afs_lookup_one_filldir,
530 .name = dentry->d_name,
531 .fid.vid = as->volume->vid
535 _enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
537 /* search the directory */
538 ret = afs_dir_iterate(dir, &cookie.ctx, key);
540 _leave(" = %d [iter]", ret);
546 _leave(" = -ENOENT [not found]");
551 _leave(" = 0 { vn=%u u=%u }", fid->vnode, fid->unique);
556 * search the directory for a name
557 * - if afs_dir_iterate_block() spots this function, it'll pass the FID
558 * uniquifier through dtype
560 static int afs_lookup_filldir(struct dir_context *ctx, const char *name,
561 int nlen, loff_t fpos, u64 ino, unsigned dtype)
563 struct afs_lookup_cookie *cookie =
564 container_of(ctx, struct afs_lookup_cookie, ctx);
567 _enter("{%s,%u},%s,%u,,%llu,%u",
568 cookie->name.name, cookie->name.len, name, nlen,
569 (unsigned long long) ino, dtype);
571 /* insanity checks first */
572 BUILD_BUG_ON(sizeof(union afs_xdr_dir_block) != 2048);
573 BUILD_BUG_ON(sizeof(union afs_xdr_dirent) != 32);
576 if (cookie->nr_fids < 50) {
577 cookie->fids[cookie->nr_fids].vnode = ino;
578 cookie->fids[cookie->nr_fids].unique = dtype;
581 } else if (cookie->name.len == nlen &&
582 memcmp(cookie->name.name, name, nlen) == 0) {
583 cookie->fids[0].vnode = ino;
584 cookie->fids[0].unique = dtype;
586 if (cookie->one_only)
590 ret = cookie->nr_fids >= 50 ? -1 : 0;
591 _leave(" = %d", ret);
596 * Do a lookup in a directory. We make use of bulk lookup to query a slew of
597 * files in one go and create inodes for them. The inode of the file we were
598 * asked for is returned.
600 static struct inode *afs_do_lookup(struct inode *dir, struct dentry *dentry,
603 struct afs_lookup_cookie *cookie;
604 struct afs_cb_interest *cbi = NULL;
605 struct afs_super_info *as = dir->i_sb->s_fs_info;
606 struct afs_iget_data data;
607 struct afs_fs_cursor fc;
608 struct afs_vnode *dvnode = AFS_FS_I(dir);
609 struct inode *inode = NULL;
612 _enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
614 cookie = kzalloc(sizeof(struct afs_lookup_cookie), GFP_KERNEL);
616 return ERR_PTR(-ENOMEM);
618 cookie->ctx.actor = afs_lookup_filldir;
619 cookie->name = dentry->d_name;
620 cookie->nr_fids = 1; /* slot 0 is saved for the fid we actually want */
622 read_seqlock_excl(&dvnode->cb_lock);
623 if (dvnode->cb_interest &&
624 dvnode->cb_interest->server &&
625 test_bit(AFS_SERVER_FL_NO_IBULK, &dvnode->cb_interest->server->flags))
626 cookie->one_only = true;
627 read_sequnlock_excl(&dvnode->cb_lock);
629 for (i = 0; i < 50; i++)
630 cookie->fids[i].vid = as->volume->vid;
632 /* search the directory */
633 ret = afs_dir_iterate(dir, &cookie->ctx, key);
635 inode = ERR_PTR(ret);
639 inode = ERR_PTR(-ENOENT);
643 /* Check to see if we already have an inode for the primary fid. */
644 data.volume = dvnode->volume;
645 data.fid = cookie->fids[0];
646 inode = ilookup5(dir->i_sb, cookie->fids[0].vnode, afs_iget5_test, &data);
650 /* Need space for examining all the selected files */
651 inode = ERR_PTR(-ENOMEM);
652 cookie->statuses = kcalloc(cookie->nr_fids, sizeof(struct afs_file_status),
654 if (!cookie->statuses)
657 cookie->callbacks = kcalloc(cookie->nr_fids, sizeof(struct afs_callback),
659 if (!cookie->callbacks)
662 /* Try FS.InlineBulkStatus first. Abort codes for the individual
663 * lookups contained therein are stored in the reply without aborting
664 * the whole operation.
666 if (cookie->one_only)
667 goto no_inline_bulk_status;
669 inode = ERR_PTR(-ERESTARTSYS);
670 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
671 while (afs_select_fileserver(&fc)) {
672 if (test_bit(AFS_SERVER_FL_NO_IBULK,
673 &fc.cbi->server->flags)) {
674 fc.ac.abort_code = RX_INVALID_OPERATION;
675 fc.ac.error = -ECONNABORTED;
678 afs_fs_inline_bulk_status(&fc,
683 cookie->nr_fids, NULL);
686 if (fc.ac.error == 0)
687 cbi = afs_get_cb_interest(fc.cbi);
688 if (fc.ac.abort_code == RX_INVALID_OPERATION)
689 set_bit(AFS_SERVER_FL_NO_IBULK, &fc.cbi->server->flags);
690 inode = ERR_PTR(afs_end_vnode_operation(&fc));
695 if (fc.ac.abort_code != RX_INVALID_OPERATION)
698 no_inline_bulk_status:
699 /* We could try FS.BulkStatus next, but this aborts the entire op if
700 * any of the lookups fails - so, for the moment, revert to
701 * FS.FetchStatus for just the primary fid.
704 inode = ERR_PTR(-ERESTARTSYS);
705 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
706 while (afs_select_fileserver(&fc)) {
707 afs_fs_fetch_status(&fc,
715 if (fc.ac.error == 0)
716 cbi = afs_get_cb_interest(fc.cbi);
717 inode = ERR_PTR(afs_end_vnode_operation(&fc));
723 for (i = 0; i < cookie->nr_fids; i++)
724 cookie->statuses[i].abort_code = 0;
727 /* Turn all the files into inodes and save the first one - which is the
728 * one we actually want.
730 if (cookie->statuses[0].abort_code != 0)
731 inode = ERR_PTR(afs_abort_to_error(cookie->statuses[0].abort_code));
733 for (i = 0; i < cookie->nr_fids; i++) {
736 if (cookie->statuses[i].abort_code != 0)
739 ti = afs_iget(dir->i_sb, key, &cookie->fids[i],
740 &cookie->statuses[i],
741 &cookie->callbacks[i],
752 afs_put_cb_interest(afs_v2net(dvnode), cbi);
753 kfree(cookie->callbacks);
755 kfree(cookie->statuses);
762 * Look up an entry in a directory with @sys substitution.
764 static struct dentry *afs_lookup_atsys(struct inode *dir, struct dentry *dentry,
767 struct afs_sysnames *subs;
768 struct afs_net *net = afs_i2net(dir);
770 char *buf, *p, *name;
775 ret = ERR_PTR(-ENOMEM);
776 p = buf = kmalloc(AFSNAMEMAX, GFP_KERNEL);
779 if (dentry->d_name.len > 4) {
780 memcpy(p, dentry->d_name.name, dentry->d_name.len - 4);
781 p += dentry->d_name.len - 4;
784 /* There is an ordered list of substitutes that we have to try. */
785 read_lock(&net->sysnames_lock);
786 subs = net->sysnames;
787 refcount_inc(&subs->usage);
788 read_unlock(&net->sysnames_lock);
790 for (i = 0; i < subs->nr; i++) {
791 name = subs->subs[i];
792 len = dentry->d_name.len - 4 + strlen(name);
793 if (len >= AFSNAMEMAX) {
794 ret = ERR_PTR(-ENAMETOOLONG);
799 ret = lookup_one_len(buf, dentry->d_parent, len);
800 if (IS_ERR(ret) || d_is_positive(ret))
805 /* We don't want to d_add() the @sys dentry here as we don't want to
806 * the cached dentry to hide changes to the sysnames list.
810 afs_put_sysnames(subs);
818 * look up an entry in a directory
820 static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
823 struct afs_vnode *dvnode = AFS_FS_I(dir);
828 _enter("{%x:%u},%p{%pd},",
829 dvnode->fid.vid, dvnode->fid.vnode, dentry, dentry);
831 ASSERTCMP(d_inode(dentry), ==, NULL);
833 if (dentry->d_name.len >= AFSNAMEMAX) {
834 _leave(" = -ENAMETOOLONG");
835 return ERR_PTR(-ENAMETOOLONG);
838 if (test_bit(AFS_VNODE_DELETED, &dvnode->flags)) {
839 _leave(" = -ESTALE");
840 return ERR_PTR(-ESTALE);
843 key = afs_request_key(dvnode->volume->cell);
845 _leave(" = %ld [key]", PTR_ERR(key));
846 return ERR_CAST(key);
849 ret = afs_validate(dvnode, key);
852 _leave(" = %d [val]", ret);
856 if (dentry->d_name.len >= 4 &&
857 dentry->d_name.name[dentry->d_name.len - 4] == '@' &&
858 dentry->d_name.name[dentry->d_name.len - 3] == 's' &&
859 dentry->d_name.name[dentry->d_name.len - 2] == 'y' &&
860 dentry->d_name.name[dentry->d_name.len - 1] == 's')
861 return afs_lookup_atsys(dir, dentry, key);
863 afs_stat_v(dvnode, n_lookup);
864 inode = afs_do_lookup(dir, dentry, key);
866 ret = PTR_ERR(inode);
867 if (ret == -ENOENT) {
868 inode = afs_try_auto_mntpt(dentry, dir);
869 if (!IS_ERR(inode)) {
874 ret = PTR_ERR(inode);
878 if (ret == -ENOENT) {
880 _leave(" = NULL [negative]");
883 _leave(" = %d [do]", ret);
886 dentry->d_fsdata = (void *)(unsigned long)dvnode->status.data_version;
888 /* instantiate the dentry */
891 _leave(" = %ld", PTR_ERR(inode));
892 return ERR_CAST(inode);
896 d_add(dentry, inode);
897 _leave(" = 0 { ino=%lu v=%u }",
898 d_inode(dentry)->i_ino,
899 d_inode(dentry)->i_generation);
905 * check that a dentry lookup hit has found a valid entry
906 * - NOTE! the hit can be a negative hit too, so we can't assume we have an
909 static int afs_d_revalidate(struct dentry *dentry, unsigned int flags)
911 struct afs_vnode *vnode, *dir;
912 struct afs_fid uninitialized_var(fid);
913 struct dentry *parent;
916 long dir_version, de_version;
919 if (flags & LOOKUP_RCU)
922 if (d_really_is_positive(dentry)) {
923 vnode = AFS_FS_I(d_inode(dentry));
924 _enter("{v={%x:%u} n=%pd fl=%lx},",
925 vnode->fid.vid, vnode->fid.vnode, dentry,
928 _enter("{neg n=%pd}", dentry);
931 key = afs_request_key(AFS_FS_S(dentry->d_sb)->volume->cell);
935 if (d_really_is_positive(dentry)) {
936 inode = d_inode(dentry);
938 vnode = AFS_FS_I(inode);
939 afs_validate(vnode, key);
940 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
945 /* lock down the parent dentry so we can peer at it */
946 parent = dget_parent(dentry);
947 dir = AFS_FS_I(d_inode(parent));
949 /* validate the parent directory */
950 afs_validate(dir, key);
952 if (test_bit(AFS_VNODE_DELETED, &dir->flags)) {
953 _debug("%pd: parent dir deleted", dentry);
957 /* We only need to invalidate a dentry if the server's copy changed
958 * behind our back. If we made the change, it's no problem. Note that
959 * on a 32-bit system, we only have 32 bits in the dentry to store the
962 dir_version = (long)dir->status.data_version;
963 de_version = (long)dentry->d_fsdata;
964 if (de_version == dir_version)
967 dir_version = (long)dir->invalid_before;
968 if (de_version - dir_version >= 0)
971 _debug("dir modified");
972 afs_stat_v(dir, n_reval);
974 /* search the directory for this vnode */
975 ret = afs_do_lookup_one(&dir->vfs_inode, dentry, &fid, key);
978 /* the filename maps to something */
979 if (d_really_is_negative(dentry))
981 inode = d_inode(dentry);
982 if (is_bad_inode(inode)) {
983 printk("kAFS: afs_d_revalidate: %pd2 has bad inode\n",
988 vnode = AFS_FS_I(inode);
990 /* if the vnode ID has changed, then the dirent points to a
992 if (fid.vnode != vnode->fid.vnode) {
993 _debug("%pd: dirent changed [%u != %u]",
999 /* if the vnode ID uniqifier has changed, then the file has
1000 * been deleted and replaced, and the original vnode ID has
1002 if (fid.unique != vnode->fid.unique) {
1003 _debug("%pd: file deleted (uq %u -> %u I:%u)",
1006 vnode->vfs_inode.i_generation);
1007 write_seqlock(&vnode->cb_lock);
1008 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1009 write_sequnlock(&vnode->cb_lock);
1015 /* the filename is unknown */
1016 _debug("%pd: dirent not found", dentry);
1017 if (d_really_is_positive(dentry))
1022 _debug("failed to iterate dir %pd: %d",
1024 goto out_bad_parent;
1028 dentry->d_fsdata = (void *)dir_version;
1031 _leave(" = 1 [valid]");
1034 /* the dirent, if it exists, now points to a different vnode */
1036 spin_lock(&dentry->d_lock);
1037 dentry->d_flags |= DCACHE_NFSFS_RENAMED;
1038 spin_unlock(&dentry->d_lock);
1041 _debug("dropping dentry %pd2", dentry);
1046 _leave(" = 0 [bad]");
1051 * allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
1053 * - called from dput() when d_count is going to 0.
1054 * - return 1 to request dentry be unhashed, 0 otherwise
1056 static int afs_d_delete(const struct dentry *dentry)
1058 _enter("%pd", dentry);
1060 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1063 if (d_really_is_positive(dentry) &&
1064 (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(d_inode(dentry))->flags) ||
1065 test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(d_inode(dentry))->flags)))
1068 _leave(" = 0 [keep]");
1072 _leave(" = 1 [zap]");
1077 * handle dentry release
1079 void afs_d_release(struct dentry *dentry)
1081 _enter("%pd", dentry);
1085 * Create a new inode for create/mkdir/symlink
1087 static void afs_vnode_new_inode(struct afs_fs_cursor *fc,
1088 struct dentry *new_dentry,
1089 struct afs_fid *newfid,
1090 struct afs_file_status *newstatus,
1091 struct afs_callback *newcb)
1093 struct afs_vnode *vnode;
1094 struct inode *inode;
1096 if (fc->ac.error < 0)
1101 inode = afs_iget(fc->vnode->vfs_inode.i_sb, fc->key,
1102 newfid, newstatus, newcb, fc->cbi);
1103 if (IS_ERR(inode)) {
1104 /* ENOMEM or EINTR at a really inconvenient time - just abandon
1105 * the new directory on the server.
1107 fc->ac.error = PTR_ERR(inode);
1111 vnode = AFS_FS_I(inode);
1112 set_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
1113 d_add(new_dentry, inode);
1117 * create a directory on an AFS filesystem
1119 static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1121 struct afs_file_status newstatus;
1122 struct afs_fs_cursor fc;
1123 struct afs_callback newcb;
1124 struct afs_vnode *dvnode = AFS_FS_I(dir);
1125 struct afs_fid newfid;
1127 u64 data_version = dvnode->status.data_version;
1132 _enter("{%x:%u},{%pd},%ho",
1133 dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1135 key = afs_request_key(dvnode->volume->cell);
1142 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1143 while (afs_select_fileserver(&fc)) {
1144 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1145 afs_fs_create(&fc, dentry->d_name.name, mode, data_version,
1146 &newfid, &newstatus, &newcb);
1149 afs_check_for_remote_deletion(&fc, fc.vnode);
1150 afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1151 afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, &newcb);
1152 ret = afs_end_vnode_operation(&fc);
1160 test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1161 afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
1162 afs_edit_dir_for_create);
1172 _leave(" = %d", ret);
1177 * Remove a subdir from a directory.
1179 static void afs_dir_remove_subdir(struct dentry *dentry)
1181 if (d_really_is_positive(dentry)) {
1182 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1184 clear_nlink(&vnode->vfs_inode);
1185 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1186 clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1187 clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags);
1192 * remove a directory from an AFS filesystem
1194 static int afs_rmdir(struct inode *dir, struct dentry *dentry)
1196 struct afs_fs_cursor fc;
1197 struct afs_vnode *dvnode = AFS_FS_I(dir);
1199 u64 data_version = dvnode->status.data_version;
1202 _enter("{%x:%u},{%pd}",
1203 dvnode->fid.vid, dvnode->fid.vnode, dentry);
1205 key = afs_request_key(dvnode->volume->cell);
1212 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1213 while (afs_select_fileserver(&fc)) {
1214 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1215 afs_fs_remove(&fc, dentry->d_name.name, true,
1219 afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1220 ret = afs_end_vnode_operation(&fc);
1222 afs_dir_remove_subdir(dentry);
1223 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1224 afs_edit_dir_remove(dvnode, &dentry->d_name,
1225 afs_edit_dir_for_rmdir);
1235 * Remove a link to a file or symlink from a directory.
1237 * If the file was not deleted due to excess hard links, the fileserver will
1238 * break the callback promise on the file - if it had one - before it returns
1239 * to us, and if it was deleted, it won't
1241 * However, if we didn't have a callback promise outstanding, or it was
1242 * outstanding on a different server, then it won't break it either...
1244 static int afs_dir_remove_link(struct dentry *dentry, struct key *key,
1245 unsigned long d_version_before,
1246 unsigned long d_version_after)
1251 /* There were no intervening changes on the server if the version
1252 * number we got back was incremented by exactly 1.
1254 dir_valid = (d_version_after == d_version_before + 1);
1256 if (d_really_is_positive(dentry)) {
1257 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
1260 drop_nlink(&vnode->vfs_inode);
1261 if (vnode->vfs_inode.i_nlink == 0) {
1262 set_bit(AFS_VNODE_DELETED, &vnode->flags);
1263 clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1267 clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
1269 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
1270 kdebug("AFS_VNODE_DELETED");
1272 ret = afs_validate(vnode, key);
1276 _debug("nlink %d [val %d]", vnode->vfs_inode.i_nlink, ret);
1283 * Remove a file or symlink from an AFS filesystem.
1285 static int afs_unlink(struct inode *dir, struct dentry *dentry)
1287 struct afs_fs_cursor fc;
1288 struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode;
1290 unsigned long d_version = (unsigned long)dentry->d_fsdata;
1291 u64 data_version = dvnode->status.data_version;
1294 _enter("{%x:%u},{%pd}",
1295 dvnode->fid.vid, dvnode->fid.vnode, dentry);
1297 if (dentry->d_name.len >= AFSNAMEMAX)
1298 return -ENAMETOOLONG;
1300 key = afs_request_key(dvnode->volume->cell);
1306 /* Try to make sure we have a callback promise on the victim. */
1307 if (d_really_is_positive(dentry)) {
1308 vnode = AFS_FS_I(d_inode(dentry));
1309 ret = afs_validate(vnode, key);
1315 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1316 while (afs_select_fileserver(&fc)) {
1317 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1318 afs_fs_remove(&fc, dentry->d_name.name, false,
1322 afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1323 ret = afs_end_vnode_operation(&fc);
1325 ret = afs_dir_remove_link(
1326 dentry, key, d_version,
1327 (unsigned long)dvnode->status.data_version);
1329 test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1330 afs_edit_dir_remove(dvnode, &dentry->d_name,
1331 afs_edit_dir_for_unlink);
1337 _leave(" = %d", ret);
1342 * create a regular file on an AFS filesystem
1344 static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1347 struct afs_fs_cursor fc;
1348 struct afs_file_status newstatus;
1349 struct afs_callback newcb;
1350 struct afs_vnode *dvnode = AFS_FS_I(dir);
1351 struct afs_fid newfid;
1353 u64 data_version = dvnode->status.data_version;
1358 _enter("{%x:%u},{%pd},%ho,",
1359 dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
1361 ret = -ENAMETOOLONG;
1362 if (dentry->d_name.len >= AFSNAMEMAX)
1365 key = afs_request_key(dvnode->volume->cell);
1372 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1373 while (afs_select_fileserver(&fc)) {
1374 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1375 afs_fs_create(&fc, dentry->d_name.name, mode, data_version,
1376 &newfid, &newstatus, &newcb);
1379 afs_check_for_remote_deletion(&fc, fc.vnode);
1380 afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1381 afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, &newcb);
1382 ret = afs_end_vnode_operation(&fc);
1389 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1390 afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
1391 afs_edit_dir_for_create);
1401 _leave(" = %d", ret);
1406 * create a hard link between files in an AFS filesystem
1408 static int afs_link(struct dentry *from, struct inode *dir,
1409 struct dentry *dentry)
1411 struct afs_fs_cursor fc;
1412 struct afs_vnode *dvnode, *vnode;
1417 vnode = AFS_FS_I(d_inode(from));
1418 dvnode = AFS_FS_I(dir);
1419 data_version = dvnode->status.data_version;
1421 _enter("{%x:%u},{%x:%u},{%pd}",
1422 vnode->fid.vid, vnode->fid.vnode,
1423 dvnode->fid.vid, dvnode->fid.vnode,
1426 ret = -ENAMETOOLONG;
1427 if (dentry->d_name.len >= AFSNAMEMAX)
1430 key = afs_request_key(dvnode->volume->cell);
1437 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1438 if (mutex_lock_interruptible_nested(&vnode->io_lock, 1) < 0) {
1439 afs_end_vnode_operation(&fc);
1443 while (afs_select_fileserver(&fc)) {
1444 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1445 fc.cb_break_2 = afs_calc_vnode_cb_break(vnode);
1446 afs_fs_link(&fc, vnode, dentry->d_name.name, data_version);
1449 afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1450 afs_vnode_commit_status(&fc, vnode, fc.cb_break_2);
1451 ihold(&vnode->vfs_inode);
1452 d_instantiate(dentry, &vnode->vfs_inode);
1454 mutex_unlock(&vnode->io_lock);
1455 ret = afs_end_vnode_operation(&fc);
1462 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1463 afs_edit_dir_add(dvnode, &dentry->d_name, &vnode->fid,
1464 afs_edit_dir_for_link);
1474 _leave(" = %d", ret);
1479 * create a symlink in an AFS filesystem
1481 static int afs_symlink(struct inode *dir, struct dentry *dentry,
1482 const char *content)
1484 struct afs_fs_cursor fc;
1485 struct afs_file_status newstatus;
1486 struct afs_vnode *dvnode = AFS_FS_I(dir);
1487 struct afs_fid newfid;
1489 u64 data_version = dvnode->status.data_version;
1492 _enter("{%x:%u},{%pd},%s",
1493 dvnode->fid.vid, dvnode->fid.vnode, dentry,
1496 ret = -ENAMETOOLONG;
1497 if (dentry->d_name.len >= AFSNAMEMAX)
1501 if (strlen(content) >= AFSPATHMAX)
1504 key = afs_request_key(dvnode->volume->cell);
1511 if (afs_begin_vnode_operation(&fc, dvnode, key)) {
1512 while (afs_select_fileserver(&fc)) {
1513 fc.cb_break = afs_calc_vnode_cb_break(dvnode);
1514 afs_fs_symlink(&fc, dentry->d_name.name,
1515 content, data_version,
1516 &newfid, &newstatus);
1519 afs_check_for_remote_deletion(&fc, fc.vnode);
1520 afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
1521 afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, NULL);
1522 ret = afs_end_vnode_operation(&fc);
1529 if (test_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1530 afs_edit_dir_add(dvnode, &dentry->d_name, &newfid,
1531 afs_edit_dir_for_symlink);
1541 _leave(" = %d", ret);
1546 * rename a file in an AFS filesystem and/or move it between directories
1548 static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
1549 struct inode *new_dir, struct dentry *new_dentry,
1552 struct afs_fs_cursor fc;
1553 struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
1555 u64 orig_data_version, new_data_version;
1556 bool new_negative = d_is_negative(new_dentry);
1562 vnode = AFS_FS_I(d_inode(old_dentry));
1563 orig_dvnode = AFS_FS_I(old_dir);
1564 new_dvnode = AFS_FS_I(new_dir);
1565 orig_data_version = orig_dvnode->status.data_version;
1566 new_data_version = new_dvnode->status.data_version;
1568 _enter("{%x:%u},{%x:%u},{%x:%u},{%pd}",
1569 orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
1570 vnode->fid.vid, vnode->fid.vnode,
1571 new_dvnode->fid.vid, new_dvnode->fid.vnode,
1574 key = afs_request_key(orig_dvnode->volume->cell);
1581 if (afs_begin_vnode_operation(&fc, orig_dvnode, key)) {
1582 if (orig_dvnode != new_dvnode) {
1583 if (mutex_lock_interruptible_nested(&new_dvnode->io_lock, 1) < 0) {
1584 afs_end_vnode_operation(&fc);
1588 while (afs_select_fileserver(&fc)) {
1589 fc.cb_break = afs_calc_vnode_cb_break(orig_dvnode);
1590 fc.cb_break_2 = afs_calc_vnode_cb_break(new_dvnode);
1591 afs_fs_rename(&fc, old_dentry->d_name.name,
1592 new_dvnode, new_dentry->d_name.name,
1593 orig_data_version, new_data_version);
1596 afs_vnode_commit_status(&fc, orig_dvnode, fc.cb_break);
1597 afs_vnode_commit_status(&fc, new_dvnode, fc.cb_break_2);
1598 if (orig_dvnode != new_dvnode)
1599 mutex_unlock(&new_dvnode->io_lock);
1600 ret = afs_end_vnode_operation(&fc);
1606 if (test_bit(AFS_VNODE_DIR_VALID, &orig_dvnode->flags))
1607 afs_edit_dir_remove(orig_dvnode, &old_dentry->d_name,
1608 afs_edit_dir_for_rename);
1610 if (!new_negative &&
1611 test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
1612 afs_edit_dir_remove(new_dvnode, &new_dentry->d_name,
1613 afs_edit_dir_for_rename);
1615 if (test_bit(AFS_VNODE_DIR_VALID, &new_dvnode->flags))
1616 afs_edit_dir_add(new_dvnode, &new_dentry->d_name,
1617 &vnode->fid, afs_edit_dir_for_rename);
1623 _leave(" = %d", ret);
1628 * Release a directory page and clean up its private state if it's not busy
1629 * - return true if the page can now be released, false if not
1631 static int afs_dir_releasepage(struct page *page, gfp_t gfp_flags)
1633 struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
1635 _enter("{{%x:%u}[%lu]}", dvnode->fid.vid, dvnode->fid.vnode, page->index);
1637 set_page_private(page, 0);
1638 ClearPagePrivate(page);
1640 /* The directory will need reloading. */
1641 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1642 afs_stat_v(dvnode, n_relpg);
1647 * invalidate part or all of a page
1648 * - release a page and clean up its private data if offset is 0 (indicating
1651 static void afs_dir_invalidatepage(struct page *page, unsigned int offset,
1652 unsigned int length)
1654 struct afs_vnode *dvnode = AFS_FS_I(page->mapping->host);
1656 _enter("{%lu},%u,%u", page->index, offset, length);
1658 BUG_ON(!PageLocked(page));
1660 /* The directory will need reloading. */
1661 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &dvnode->flags))
1662 afs_stat_v(dvnode, n_inval);
1664 /* we clean up only if the entire page is being invalidated */
1665 if (offset == 0 && length == PAGE_SIZE) {
1666 set_page_private(page, 0);
1667 ClearPagePrivate(page);