1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/spinlock.h>
4 #include <linux/fs_struct.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
18 * Ceph MDS operations are specified in terms of a base ino and
19 * relative path. Thus, the client can specify an operation on a
20 * specific inode (e.g., a getattr due to fstat(2)), or as a path
21 * relative to, say, the root directory.
23 * Normally, we limit ourselves to strict inode ops (no path component)
24 * or dentry operations (a single path component relative to an ino). The
25 * exception to this is open_root_dentry(), which will open the mount
29 const struct inode_operations ceph_dir_iops;
30 const struct file_operations ceph_dir_fops;
31 const struct dentry_operations ceph_dentry_ops;
34 * Initialize ceph dentry state.
36 int ceph_init_dentry(struct dentry *dentry)
38 struct ceph_dentry_info *di;
43 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
45 return -ENOMEM; /* oh well */
47 spin_lock(&dentry->d_lock);
48 if (dentry->d_fsdata) {
50 kmem_cache_free(ceph_dentry_cachep, di);
54 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
55 d_set_d_op(dentry, &ceph_dentry_ops);
56 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
57 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
59 d_set_d_op(dentry, &ceph_snap_dentry_ops);
62 di->lease_session = NULL;
63 dentry->d_time = jiffies;
64 /* avoid reordering d_fsdata setup so that the check above is safe */
66 dentry->d_fsdata = di;
67 ceph_dentry_lru_add(dentry);
69 spin_unlock(&dentry->d_lock);
73 struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
75 struct inode *inode = NULL;
80 spin_lock(&dentry->d_lock);
81 if (!IS_ROOT(dentry)) {
82 inode = dentry->d_parent->d_inode;
85 spin_unlock(&dentry->d_lock);
91 * for readdir, we encode the directory frag and offset within that
94 static unsigned fpos_frag(loff_t p)
98 static unsigned fpos_off(loff_t p)
100 return p & 0xffffffff;
103 static int fpos_cmp(loff_t l, loff_t r)
105 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
108 return (int)(fpos_off(l) - fpos_off(r));
112 * When possible, we try to satisfy a readdir by peeking at the
113 * dcache. We make this work by carefully ordering dentries on
114 * d_u.d_child when we initially get results back from the MDS, and
115 * falling back to a "normal" sync readdir if any dentries in the dir
118 * Complete dir indicates that we have all dentries in the dir. It is
119 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
120 * the MDS if/when the directory is modified).
122 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
125 struct ceph_file_info *fi = file->private_data;
126 struct dentry *parent = file->f_dentry;
127 struct inode *dir = parent->d_inode;
129 struct dentry *dentry, *last;
130 struct ceph_dentry_info *di;
133 /* claim ref on last dentry we returned */
137 dout("__dcache_readdir %p v%u at %llu (last %p)\n",
138 dir, shared_gen, ctx->pos, last);
140 spin_lock(&parent->d_lock);
142 /* start at beginning? */
143 if (ctx->pos == 2 || last == NULL ||
144 ctx->pos < ceph_dentry(last)->offset) {
145 if (list_empty(&parent->d_subdirs))
147 p = parent->d_subdirs.prev;
148 dout(" initial p %p/%p\n", p->prev, p->next);
150 p = last->d_u.d_child.prev;
154 dentry = list_entry(p, struct dentry, d_u.d_child);
155 di = ceph_dentry(dentry);
157 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
158 d_unhashed(dentry) ? "!hashed" : "hashed",
159 parent->d_subdirs.prev, parent->d_subdirs.next);
160 if (p == &parent->d_subdirs) {
161 fi->flags |= CEPH_F_ATEND;
164 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
165 if (di->lease_shared_gen == shared_gen &&
166 !d_unhashed(dentry) && dentry->d_inode &&
167 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
168 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
169 fpos_cmp(ctx->pos, di->offset) <= 0)
171 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
172 dentry->d_name.len, dentry->d_name.name, di->offset,
173 ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
174 !dentry->d_inode ? " null" : "");
175 spin_unlock(&dentry->d_lock);
177 dentry = list_entry(p, struct dentry, d_u.d_child);
178 di = ceph_dentry(dentry);
182 spin_unlock(&dentry->d_lock);
183 spin_unlock(&parent->d_lock);
185 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos,
186 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
187 ctx->pos = di->offset;
188 if (!dir_emit(ctx, dentry->d_name.name,
190 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
191 dentry->d_inode->i_mode >> 12)) {
193 /* remember our position */
195 fi->next_offset = fpos_off(di->offset);
207 /* make sure a dentry wasn't dropped while we didn't have parent lock */
208 if (!ceph_dir_is_complete(dir)) {
209 dout(" lost dir complete on %p; falling back to mds\n", dir);
214 spin_lock(&parent->d_lock);
215 p = p->prev; /* advance to next dentry */
219 spin_unlock(&parent->d_lock);
227 * make note of the last dentry we read, so we can
228 * continue at the same lexicographical point,
229 * regardless of what dir changes take place on the
232 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
235 kfree(fi->last_name);
236 fi->last_name = kmalloc(len+1, GFP_NOFS);
239 memcpy(fi->last_name, name, len);
240 fi->last_name[len] = 0;
241 dout("note_last_dentry '%s'\n", fi->last_name);
245 static int ceph_readdir(struct file *file, struct dir_context *ctx)
247 struct ceph_file_info *fi = file->private_data;
248 struct inode *inode = file_inode(file);
249 struct ceph_inode_info *ci = ceph_inode(inode);
250 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
251 struct ceph_mds_client *mdsc = fsc->mdsc;
252 unsigned frag = fpos_frag(ctx->pos);
253 int off = fpos_off(ctx->pos);
256 struct ceph_mds_reply_info_parsed *rinfo;
258 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
259 if (fi->flags & CEPH_F_ATEND)
262 /* always start with . and .. */
264 /* note dir version at start of readdir so we can tell
265 * if any dentries get dropped */
266 fi->dir_release_count = atomic_read(&ci->i_release_count);
268 dout("readdir off 0 -> '.'\n");
269 if (!dir_emit(ctx, ".", 1,
270 ceph_translate_ino(inode->i_sb, inode->i_ino),
271 inode->i_mode >> 12))
277 ino_t ino = parent_ino(file->f_dentry);
278 dout("readdir off 1 -> '..'\n");
279 if (!dir_emit(ctx, "..", 2,
280 ceph_translate_ino(inode->i_sb, ino),
281 inode->i_mode >> 12))
287 /* can we use the dcache? */
288 spin_lock(&ci->i_ceph_lock);
289 if ((ctx->pos == 2 || fi->dentry) &&
290 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
291 ceph_snap(inode) != CEPH_SNAPDIR &&
292 __ceph_dir_is_complete(ci) &&
293 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
294 u32 shared_gen = ci->i_shared_gen;
295 spin_unlock(&ci->i_ceph_lock);
296 err = __dcache_readdir(file, ctx, shared_gen);
300 spin_unlock(&ci->i_ceph_lock);
303 err = note_last_dentry(fi, fi->dentry->d_name.name,
304 fi->dentry->d_name.len);
311 /* proceed with a normal readdir */
314 /* do we have the correct frag content buffered? */
315 if (fi->frag != frag || fi->last_readdir == NULL) {
316 struct ceph_mds_request *req;
317 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
318 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
320 /* discard old result, if any */
321 if (fi->last_readdir) {
322 ceph_mdsc_put_request(fi->last_readdir);
323 fi->last_readdir = NULL;
326 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
327 ceph_vinop(inode), frag, fi->last_name);
328 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
331 err = ceph_alloc_readdir_reply_buffer(req, inode);
333 ceph_mdsc_put_request(req);
336 req->r_inode = inode;
338 req->r_dentry = dget(file->f_dentry);
339 /* hints to request -> mds selection code */
340 req->r_direct_mode = USE_AUTH_MDS;
341 req->r_direct_hash = ceph_frag_value(frag);
342 req->r_direct_is_hash = true;
343 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
344 req->r_readdir_offset = fi->next_offset;
345 req->r_args.readdir.frag = cpu_to_le32(frag);
346 err = ceph_mdsc_do_request(mdsc, NULL, req);
348 ceph_mdsc_put_request(req);
351 dout("readdir got and parsed readdir result=%d"
352 " on frag %x, end=%d, complete=%d\n", err, frag,
353 (int)req->r_reply_info.dir_end,
354 (int)req->r_reply_info.dir_complete);
356 if (!req->r_did_prepopulate) {
357 dout("readdir !did_prepopulate");
358 /* preclude from marking dir complete */
359 fi->dir_release_count--;
362 /* note next offset and last dentry name */
363 rinfo = &req->r_reply_info;
364 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
365 frag = le32_to_cpu(rinfo->dir_dir->frag);
366 if (ceph_frag_is_leftmost(frag))
370 off = fi->next_offset;
373 fi->offset = fi->next_offset;
374 fi->last_readdir = req;
376 if (req->r_reply_info.dir_end) {
377 kfree(fi->last_name);
378 fi->last_name = NULL;
379 if (ceph_frag_is_rightmost(frag))
384 err = note_last_dentry(fi,
385 rinfo->dir_dname[rinfo->dir_nr-1],
386 rinfo->dir_dname_len[rinfo->dir_nr-1]);
389 fi->next_offset += rinfo->dir_nr;
393 rinfo = &fi->last_readdir->r_reply_info;
394 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
395 rinfo->dir_nr, off, fi->offset);
397 ctx->pos = ceph_make_fpos(frag, off);
398 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
399 struct ceph_mds_reply_inode *in =
400 rinfo->dir_in[off - fi->offset].in;
401 struct ceph_vino vino;
404 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
405 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
406 rinfo->dir_dname_len[off - fi->offset],
407 rinfo->dir_dname[off - fi->offset], in);
409 ftype = le32_to_cpu(in->mode) >> 12;
410 vino.ino = le64_to_cpu(in->ino);
411 vino.snap = le64_to_cpu(in->snapid);
412 ino = ceph_vino_to_ino(vino);
414 rinfo->dir_dname[off - fi->offset],
415 rinfo->dir_dname_len[off - fi->offset],
416 ceph_translate_ino(inode->i_sb, ino), ftype)) {
417 dout("filldir stopping us...\n");
425 ceph_mdsc_put_request(fi->last_readdir);
426 fi->last_readdir = NULL;
431 if (!ceph_frag_is_rightmost(frag)) {
432 frag = ceph_frag_next(frag);
434 ctx->pos = ceph_make_fpos(frag, off);
435 dout("readdir next frag is %x\n", frag);
438 fi->flags |= CEPH_F_ATEND;
441 * if dir_release_count still matches the dir, no dentries
442 * were released during the whole readdir, and we should have
443 * the complete dir contents in our cache.
445 spin_lock(&ci->i_ceph_lock);
446 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
447 dout(" marking %p complete\n", inode);
448 __ceph_dir_set_complete(ci, fi->dir_release_count);
449 ci->i_max_offset = ctx->pos;
451 spin_unlock(&ci->i_ceph_lock);
453 dout("readdir %p file %p done.\n", inode, file);
457 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
459 if (fi->last_readdir) {
460 ceph_mdsc_put_request(fi->last_readdir);
461 fi->last_readdir = NULL;
463 kfree(fi->last_name);
464 fi->last_name = NULL;
465 if (ceph_frag_is_leftmost(frag))
466 fi->next_offset = 2; /* compensate for . and .. */
473 fi->flags &= ~CEPH_F_ATEND;
476 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
478 struct ceph_file_info *fi = file->private_data;
479 struct inode *inode = file->f_mapping->host;
480 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
483 mutex_lock(&inode->i_mutex);
487 offset += inode->i_size + 2; /* FIXME */
490 offset += file->f_pos;
498 if (offset != file->f_pos) {
499 file->f_pos = offset;
501 fi->flags &= ~CEPH_F_ATEND;
506 * discard buffered readdir content on seekdir(0), or
507 * seek to new frag, or seek prior to current chunk.
510 fpos_frag(offset) != fi->frag ||
511 fpos_off(offset) < fi->offset) {
512 dout("dir_llseek dropping %p content\n", file);
513 reset_readdir(fi, fpos_frag(offset));
516 /* bump dir_release_count if we did a forward seek */
517 if (fpos_cmp(offset, old_offset) > 0)
518 fi->dir_release_count--;
521 mutex_unlock(&inode->i_mutex);
526 * Handle lookups for the hidden .snap directory.
528 int ceph_handle_snapdir(struct ceph_mds_request *req,
529 struct dentry *dentry, int err)
531 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
532 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
535 if (err == -ENOENT &&
536 ceph_snap(parent) == CEPH_NOSNAP &&
537 strcmp(dentry->d_name.name,
538 fsc->mount_options->snapdir_name) == 0) {
539 struct inode *inode = ceph_get_snapdir(parent);
540 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
541 dentry, dentry->d_name.len, dentry->d_name.name, inode);
542 BUG_ON(!d_unhashed(dentry));
543 d_add(dentry, inode);
550 * Figure out final result of a lookup/open request.
552 * Mainly, make sure we return the final req->r_dentry (if it already
553 * existed) in place of the original VFS-provided dentry when they
556 * Gracefully handle the case where the MDS replies with -ENOENT and
557 * no trace (which it may do, at its discretion, e.g., if it doesn't
558 * care to issue a lease on the negative dentry).
560 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
561 struct dentry *dentry, int err)
563 if (err == -ENOENT) {
566 if (!req->r_reply_info.head->is_dentry) {
567 dout("ENOENT and no trace, dentry %p inode %p\n",
568 dentry, dentry->d_inode);
569 if (dentry->d_inode) {
578 dentry = ERR_PTR(err);
579 else if (dentry != req->r_dentry)
580 dentry = dget(req->r_dentry); /* we got spliced */
586 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
588 return ceph_ino(inode) == CEPH_INO_ROOT &&
589 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
593 * Look up a single dir entry. If there is a lookup intent, inform
594 * the MDS so that it gets our 'caps wanted' value in a single op.
596 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
599 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
600 struct ceph_mds_client *mdsc = fsc->mdsc;
601 struct ceph_mds_request *req;
605 dout("lookup %p dentry %p '%.*s'\n",
606 dir, dentry, dentry->d_name.len, dentry->d_name.name);
608 if (dentry->d_name.len > NAME_MAX)
609 return ERR_PTR(-ENAMETOOLONG);
611 err = ceph_init_dentry(dentry);
615 /* can we conclude ENOENT locally? */
616 if (dentry->d_inode == NULL) {
617 struct ceph_inode_info *ci = ceph_inode(dir);
618 struct ceph_dentry_info *di = ceph_dentry(dentry);
620 spin_lock(&ci->i_ceph_lock);
621 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
622 if (strncmp(dentry->d_name.name,
623 fsc->mount_options->snapdir_name,
624 dentry->d_name.len) &&
625 !is_root_ceph_dentry(dir, dentry) &&
626 __ceph_dir_is_complete(ci) &&
627 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
628 spin_unlock(&ci->i_ceph_lock);
629 dout(" dir %p complete, -ENOENT\n", dir);
631 di->lease_shared_gen = ci->i_shared_gen;
634 spin_unlock(&ci->i_ceph_lock);
637 op = ceph_snap(dir) == CEPH_SNAPDIR ?
638 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
639 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
641 return ERR_CAST(req);
642 req->r_dentry = dget(dentry);
644 /* we only need inode linkage */
645 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
646 req->r_locked_dir = dir;
647 err = ceph_mdsc_do_request(mdsc, NULL, req);
648 err = ceph_handle_snapdir(req, dentry, err);
649 dentry = ceph_finish_lookup(req, dentry, err);
650 ceph_mdsc_put_request(req); /* will dput(dentry) */
651 dout("lookup result=%p\n", dentry);
656 * If we do a create but get no trace back from the MDS, follow up with
657 * a lookup (the VFS expects us to link up the provided dentry).
659 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
661 struct dentry *result = ceph_lookup(dir, dentry, 0);
663 if (result && !IS_ERR(result)) {
665 * We created the item, then did a lookup, and found
666 * it was already linked to another inode we already
667 * had in our cache (and thus got spliced). Link our
668 * dentry to that inode, but don't hash it, just in
669 * case the VFS wants to dereference it.
671 BUG_ON(!result->d_inode);
672 d_instantiate(dentry, result->d_inode);
675 return PTR_ERR(result);
678 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
679 umode_t mode, dev_t rdev)
681 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
682 struct ceph_mds_client *mdsc = fsc->mdsc;
683 struct ceph_mds_request *req;
686 if (ceph_snap(dir) != CEPH_NOSNAP)
689 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
690 dir, dentry, mode, rdev);
691 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
696 req->r_dentry = dget(dentry);
698 req->r_locked_dir = dir;
699 req->r_args.mknod.mode = cpu_to_le32(mode);
700 req->r_args.mknod.rdev = cpu_to_le32(rdev);
701 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
702 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
703 err = ceph_mdsc_do_request(mdsc, dir, req);
704 if (!err && !req->r_reply_info.head->is_dentry)
705 err = ceph_handle_notrace_create(dir, dentry);
706 ceph_mdsc_put_request(req);
709 ceph_init_acl(dentry, dentry->d_inode, dir);
715 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
718 return ceph_mknod(dir, dentry, mode, 0);
721 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
724 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
725 struct ceph_mds_client *mdsc = fsc->mdsc;
726 struct ceph_mds_request *req;
729 if (ceph_snap(dir) != CEPH_NOSNAP)
732 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
733 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
738 req->r_dentry = dget(dentry);
740 req->r_path2 = kstrdup(dest, GFP_NOFS);
741 req->r_locked_dir = dir;
742 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
743 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
744 err = ceph_mdsc_do_request(mdsc, dir, req);
745 if (!err && !req->r_reply_info.head->is_dentry)
746 err = ceph_handle_notrace_create(dir, dentry);
747 ceph_mdsc_put_request(req);
749 ceph_init_acl(dentry, dentry->d_inode, dir);
755 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
757 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
758 struct ceph_mds_client *mdsc = fsc->mdsc;
759 struct ceph_mds_request *req;
763 if (ceph_snap(dir) == CEPH_SNAPDIR) {
764 /* mkdir .snap/foo is a MKSNAP */
765 op = CEPH_MDS_OP_MKSNAP;
766 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
767 dentry->d_name.len, dentry->d_name.name, dentry);
768 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
769 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
770 op = CEPH_MDS_OP_MKDIR;
774 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
780 req->r_dentry = dget(dentry);
782 req->r_locked_dir = dir;
783 req->r_args.mkdir.mode = cpu_to_le32(mode);
784 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
785 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
786 err = ceph_mdsc_do_request(mdsc, dir, req);
787 if (!err && !req->r_reply_info.head->is_dentry)
788 err = ceph_handle_notrace_create(dir, dentry);
789 ceph_mdsc_put_request(req);
792 ceph_init_acl(dentry, dentry->d_inode, dir);
798 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
799 struct dentry *dentry)
801 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
802 struct ceph_mds_client *mdsc = fsc->mdsc;
803 struct ceph_mds_request *req;
806 if (ceph_snap(dir) != CEPH_NOSNAP)
809 dout("link in dir %p old_dentry %p dentry %p\n", dir,
811 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
816 req->r_dentry = dget(dentry);
818 req->r_old_dentry = dget(old_dentry);
819 req->r_locked_dir = dir;
820 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
821 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
822 /* release LINK_SHARED on source inode (mds will lock it) */
823 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
824 err = ceph_mdsc_do_request(mdsc, dir, req);
827 } else if (!req->r_reply_info.head->is_dentry) {
828 ihold(old_dentry->d_inode);
829 d_instantiate(dentry, old_dentry->d_inode);
831 ceph_mdsc_put_request(req);
836 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
837 * looks like the link count will hit 0, drop any other caps (other
838 * than PIN) we don't specifically want (due to the file still being
841 static int drop_caps_for_unlink(struct inode *inode)
843 struct ceph_inode_info *ci = ceph_inode(inode);
844 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
846 spin_lock(&ci->i_ceph_lock);
847 if (inode->i_nlink == 1) {
848 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
849 ci->i_ceph_flags |= CEPH_I_NODELAY;
851 spin_unlock(&ci->i_ceph_lock);
856 * rmdir and unlink are differ only by the metadata op code
858 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
860 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
861 struct ceph_mds_client *mdsc = fsc->mdsc;
862 struct inode *inode = dentry->d_inode;
863 struct ceph_mds_request *req;
867 if (ceph_snap(dir) == CEPH_SNAPDIR) {
868 /* rmdir .snap/foo is RMSNAP */
869 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
870 dentry->d_name.name, dentry);
871 op = CEPH_MDS_OP_RMSNAP;
872 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
873 dout("unlink/rmdir dir %p dn %p inode %p\n",
875 op = S_ISDIR(dentry->d_inode->i_mode) ?
876 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
879 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
884 req->r_dentry = dget(dentry);
886 req->r_locked_dir = dir;
887 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
888 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
889 req->r_inode_drop = drop_caps_for_unlink(inode);
890 err = ceph_mdsc_do_request(mdsc, dir, req);
891 if (!err && !req->r_reply_info.head->is_dentry)
893 ceph_mdsc_put_request(req);
898 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
899 struct inode *new_dir, struct dentry *new_dentry)
901 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
902 struct ceph_mds_client *mdsc = fsc->mdsc;
903 struct ceph_mds_request *req;
906 if (ceph_snap(old_dir) != ceph_snap(new_dir))
908 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
909 ceph_snap(new_dir) != CEPH_NOSNAP)
911 dout("rename dir %p dentry %p to dir %p dentry %p\n",
912 old_dir, old_dentry, new_dir, new_dentry);
913 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
917 req->r_dentry = dget(new_dentry);
919 req->r_old_dentry = dget(old_dentry);
920 req->r_old_dentry_dir = old_dir;
921 req->r_locked_dir = new_dir;
922 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
923 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
924 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
925 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
926 /* release LINK_RDCACHE on source inode (mds will lock it) */
927 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
928 if (new_dentry->d_inode)
929 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
930 err = ceph_mdsc_do_request(mdsc, old_dir, req);
931 if (!err && !req->r_reply_info.head->is_dentry) {
933 * Normally d_move() is done by fill_trace (called by
934 * do_request, above). If there is no trace, we need
938 /* d_move screws up d_subdirs order */
939 ceph_dir_clear_complete(new_dir);
941 d_move(old_dentry, new_dentry);
943 /* ensure target dentry is invalidated, despite
944 rehashing bug in vfs_rename_dir */
945 ceph_invalidate_dentry_lease(new_dentry);
947 ceph_mdsc_put_request(req);
952 * Ensure a dentry lease will no longer revalidate.
954 void ceph_invalidate_dentry_lease(struct dentry *dentry)
956 spin_lock(&dentry->d_lock);
957 dentry->d_time = jiffies;
958 ceph_dentry(dentry)->lease_shared_gen = 0;
959 spin_unlock(&dentry->d_lock);
963 * Check if dentry lease is valid. If not, delete the lease. Try to
964 * renew if the least is more than half up.
966 static int dentry_lease_is_valid(struct dentry *dentry)
968 struct ceph_dentry_info *di;
969 struct ceph_mds_session *s;
973 struct ceph_mds_session *session = NULL;
974 struct inode *dir = NULL;
977 spin_lock(&dentry->d_lock);
978 di = ceph_dentry(dentry);
979 if (di->lease_session) {
980 s = di->lease_session;
981 spin_lock(&s->s_gen_ttl_lock);
984 spin_unlock(&s->s_gen_ttl_lock);
986 if (di->lease_gen == gen &&
987 time_before(jiffies, dentry->d_time) &&
988 time_before(jiffies, ttl)) {
990 if (di->lease_renew_after &&
991 time_after(jiffies, di->lease_renew_after)) {
992 /* we should renew */
993 dir = dentry->d_parent->d_inode;
994 session = ceph_get_mds_session(s);
996 di->lease_renew_after = 0;
997 di->lease_renew_from = jiffies;
1001 spin_unlock(&dentry->d_lock);
1004 ceph_mdsc_lease_send_msg(session, dir, dentry,
1005 CEPH_MDS_LEASE_RENEW, seq);
1006 ceph_put_mds_session(session);
1008 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1013 * Check if directory-wide content lease/cap is valid.
1015 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1017 struct ceph_inode_info *ci = ceph_inode(dir);
1018 struct ceph_dentry_info *di = ceph_dentry(dentry);
1021 spin_lock(&ci->i_ceph_lock);
1022 if (ci->i_shared_gen == di->lease_shared_gen)
1023 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1024 spin_unlock(&ci->i_ceph_lock);
1025 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1026 dir, (unsigned)ci->i_shared_gen, dentry,
1027 (unsigned)di->lease_shared_gen, valid);
1032 * Check if cached dentry can be trusted.
1034 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1039 if (flags & LOOKUP_RCU)
1042 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1043 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1044 ceph_dentry(dentry)->offset);
1046 dir = ceph_get_dentry_parent_inode(dentry);
1048 /* always trust cached snapped dentries, snapdir dentry */
1049 if (ceph_snap(dir) != CEPH_NOSNAP) {
1050 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1051 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1053 } else if (dentry->d_inode &&
1054 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1056 } else if (dentry_lease_is_valid(dentry) ||
1057 dir_lease_is_valid(dir, dentry)) {
1058 if (dentry->d_inode)
1059 valid = ceph_is_any_caps(dentry->d_inode);
1064 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1066 ceph_dentry_lru_touch(dentry);
1068 ceph_dir_clear_complete(dir);
1076 * Release our ceph_dentry_info.
1078 static void ceph_d_release(struct dentry *dentry)
1080 struct ceph_dentry_info *di = ceph_dentry(dentry);
1082 dout("d_release %p\n", dentry);
1083 ceph_dentry_lru_del(dentry);
1084 if (di->lease_session)
1085 ceph_put_mds_session(di->lease_session);
1086 kmem_cache_free(ceph_dentry_cachep, di);
1087 dentry->d_fsdata = NULL;
1090 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1094 * Eventually, we'll want to revalidate snapped metadata
1095 * too... probably...
1101 * When the VFS prunes a dentry from the cache, we need to clear the
1102 * complete flag on the parent directory.
1104 * Called under dentry->d_lock.
1106 static void ceph_d_prune(struct dentry *dentry)
1108 dout("ceph_d_prune %p\n", dentry);
1110 /* do we have a valid parent? */
1111 if (IS_ROOT(dentry))
1114 /* if we are not hashed, we don't affect dir's completeness */
1115 if (d_unhashed(dentry))
1119 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1120 * cleared until d_release
1122 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1126 * read() on a dir. This weird interface hack only works if mounted
1127 * with '-o dirstat'.
1129 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1132 struct ceph_file_info *cf = file->private_data;
1133 struct inode *inode = file_inode(file);
1134 struct ceph_inode_info *ci = ceph_inode(inode);
1136 const int bufsize = 1024;
1138 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1141 if (!cf->dir_info) {
1142 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1146 snprintf(cf->dir_info, bufsize,
1149 " subdirs: %20lld\n"
1150 "rentries: %20lld\n"
1152 " rsubdirs: %20lld\n"
1154 "rctime: %10ld.%09ld\n",
1155 ci->i_files + ci->i_subdirs,
1158 ci->i_rfiles + ci->i_rsubdirs,
1162 (long)ci->i_rctime.tv_sec,
1163 (long)ci->i_rctime.tv_nsec);
1166 if (*ppos >= cf->dir_info_len)
1168 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1169 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1172 *ppos += (size - left);
1177 * an fsync() on a dir will wait for any uncommitted directory
1178 * operations to commit.
1180 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1183 struct inode *inode = file_inode(file);
1184 struct ceph_inode_info *ci = ceph_inode(inode);
1185 struct list_head *head = &ci->i_unsafe_dirops;
1186 struct ceph_mds_request *req;
1190 dout("dir_fsync %p\n", inode);
1191 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1194 mutex_lock(&inode->i_mutex);
1196 spin_lock(&ci->i_unsafe_lock);
1197 if (list_empty(head))
1200 req = list_entry(head->prev,
1201 struct ceph_mds_request, r_unsafe_dir_item);
1202 last_tid = req->r_tid;
1205 ceph_mdsc_get_request(req);
1206 spin_unlock(&ci->i_unsafe_lock);
1208 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1209 inode, req->r_tid, last_tid);
1210 if (req->r_timeout) {
1211 ret = wait_for_completion_timeout(
1212 &req->r_safe_completion, req->r_timeout);
1216 ret = -EIO; /* timed out */
1218 wait_for_completion(&req->r_safe_completion);
1220 ceph_mdsc_put_request(req);
1222 spin_lock(&ci->i_unsafe_lock);
1223 if (ret || list_empty(head))
1225 req = list_entry(head->next,
1226 struct ceph_mds_request, r_unsafe_dir_item);
1227 } while (req->r_tid < last_tid);
1229 spin_unlock(&ci->i_unsafe_lock);
1230 mutex_unlock(&inode->i_mutex);
1236 * We maintain a private dentry LRU.
1238 * FIXME: this needs to be changed to a per-mds lru to be useful.
1240 void ceph_dentry_lru_add(struct dentry *dn)
1242 struct ceph_dentry_info *di = ceph_dentry(dn);
1243 struct ceph_mds_client *mdsc;
1245 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1246 dn->d_name.len, dn->d_name.name);
1247 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1248 spin_lock(&mdsc->dentry_lru_lock);
1249 list_add_tail(&di->lru, &mdsc->dentry_lru);
1251 spin_unlock(&mdsc->dentry_lru_lock);
1254 void ceph_dentry_lru_touch(struct dentry *dn)
1256 struct ceph_dentry_info *di = ceph_dentry(dn);
1257 struct ceph_mds_client *mdsc;
1259 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1260 dn->d_name.len, dn->d_name.name, di->offset);
1261 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1262 spin_lock(&mdsc->dentry_lru_lock);
1263 list_move_tail(&di->lru, &mdsc->dentry_lru);
1264 spin_unlock(&mdsc->dentry_lru_lock);
1267 void ceph_dentry_lru_del(struct dentry *dn)
1269 struct ceph_dentry_info *di = ceph_dentry(dn);
1270 struct ceph_mds_client *mdsc;
1272 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1273 dn->d_name.len, dn->d_name.name);
1274 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1275 spin_lock(&mdsc->dentry_lru_lock);
1276 list_del_init(&di->lru);
1278 spin_unlock(&mdsc->dentry_lru_lock);
1282 * Return name hash for a given dentry. This is dependent on
1283 * the parent directory's hash function.
1285 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1287 struct ceph_inode_info *dci = ceph_inode(dir);
1289 switch (dci->i_dir_layout.dl_dir_hash) {
1290 case 0: /* for backward compat */
1291 case CEPH_STR_HASH_LINUX:
1292 return dn->d_name.hash;
1295 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1296 dn->d_name.name, dn->d_name.len);
1300 const struct file_operations ceph_dir_fops = {
1301 .read = ceph_read_dir,
1302 .iterate = ceph_readdir,
1303 .llseek = ceph_dir_llseek,
1305 .release = ceph_release,
1306 .unlocked_ioctl = ceph_ioctl,
1307 .fsync = ceph_dir_fsync,
1310 const struct inode_operations ceph_dir_iops = {
1311 .lookup = ceph_lookup,
1312 .permission = ceph_permission,
1313 .getattr = ceph_getattr,
1314 .setattr = ceph_setattr,
1315 .setxattr = ceph_setxattr,
1316 .getxattr = ceph_getxattr,
1317 .listxattr = ceph_listxattr,
1318 .removexattr = ceph_removexattr,
1319 .get_acl = ceph_get_acl,
1320 .set_acl = ceph_set_acl,
1321 .mknod = ceph_mknod,
1322 .symlink = ceph_symlink,
1323 .mkdir = ceph_mkdir,
1325 .unlink = ceph_unlink,
1326 .rmdir = ceph_unlink,
1327 .rename = ceph_rename,
1328 .create = ceph_create,
1329 .atomic_open = ceph_atomic_open,
1332 const struct dentry_operations ceph_dentry_ops = {
1333 .d_revalidate = ceph_d_revalidate,
1334 .d_release = ceph_d_release,
1335 .d_prune = ceph_d_prune,
1338 const struct dentry_operations ceph_snapdir_dentry_ops = {
1339 .d_revalidate = ceph_snapdir_d_revalidate,
1340 .d_release = ceph_d_release,
1343 const struct dentry_operations ceph_snap_dentry_ops = {
1344 .d_release = ceph_d_release,
1345 .d_prune = ceph_d_prune,