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 if (dentry->d_parent == NULL || /* nfs fh_to_dentry */
44 ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
45 dentry->d_op = &ceph_dentry_ops;
46 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
47 dentry->d_op = &ceph_snapdir_dentry_ops;
49 dentry->d_op = &ceph_snap_dentry_ops;
51 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
53 return -ENOMEM; /* oh well */
55 spin_lock(&dentry->d_lock);
56 if (dentry->d_fsdata) {
58 kmem_cache_free(ceph_dentry_cachep, di);
62 di->lease_session = NULL;
63 dentry->d_fsdata = di;
64 dentry->d_time = jiffies;
65 ceph_dentry_lru_add(dentry);
67 spin_unlock(&dentry->d_lock);
74 * for readdir, we encode the directory frag and offset within that
77 static unsigned fpos_frag(loff_t p)
81 static unsigned fpos_off(loff_t p)
83 return p & 0xffffffff;
87 * When possible, we try to satisfy a readdir by peeking at the
88 * dcache. We make this work by carefully ordering dentries on
89 * d_u.d_child when we initially get results back from the MDS, and
90 * falling back to a "normal" sync readdir if any dentries in the dir
93 * I_COMPLETE tells indicates we have all dentries in the dir. It is
94 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
95 * the MDS if/when the directory is modified).
97 static int __dcache_readdir(struct file *filp,
98 void *dirent, filldir_t filldir)
100 struct ceph_file_info *fi = filp->private_data;
101 struct dentry *parent = filp->f_dentry;
102 struct inode *dir = parent->d_inode;
104 struct dentry *dentry, *last;
105 struct ceph_dentry_info *di;
108 /* claim ref on last dentry we returned */
112 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
115 spin_lock(&dcache_lock);
117 /* start at beginning? */
118 if (filp->f_pos == 2 || last == NULL ||
119 filp->f_pos < ceph_dentry(last)->offset) {
120 if (list_empty(&parent->d_subdirs))
122 p = parent->d_subdirs.prev;
123 dout(" initial p %p/%p\n", p->prev, p->next);
125 p = last->d_u.d_child.prev;
129 dentry = list_entry(p, struct dentry, d_u.d_child);
130 di = ceph_dentry(dentry);
132 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
133 d_unhashed(dentry) ? "!hashed" : "hashed",
134 parent->d_subdirs.prev, parent->d_subdirs.next);
135 if (p == &parent->d_subdirs) {
139 if (!d_unhashed(dentry) && dentry->d_inode &&
140 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
141 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
142 filp->f_pos <= di->offset)
144 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
145 dentry->d_name.len, dentry->d_name.name, di->offset,
146 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
147 !dentry->d_inode ? " null" : "");
149 dentry = list_entry(p, struct dentry, d_u.d_child);
150 di = ceph_dentry(dentry);
153 spin_lock(&dentry->d_lock);
155 spin_unlock(&dentry->d_lock);
156 spin_unlock(&dcache_lock);
158 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
159 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
160 filp->f_pos = di->offset;
161 err = filldir(dirent, dentry->d_name.name,
162 dentry->d_name.len, di->offset,
163 dentry->d_inode->i_ino,
164 dentry->d_inode->i_mode >> 12);
168 /* remember our position */
170 fi->next_offset = di->offset;
182 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
183 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
184 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
189 spin_lock(&dcache_lock);
190 p = p->prev; /* advance to next dentry */
194 spin_unlock(&dcache_lock);
202 * make note of the last dentry we read, so we can
203 * continue at the same lexicographical point,
204 * regardless of what dir changes take place on the
207 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
210 kfree(fi->last_name);
211 fi->last_name = kmalloc(len+1, GFP_NOFS);
214 memcpy(fi->last_name, name, len);
215 fi->last_name[len] = 0;
216 dout("note_last_dentry '%s'\n", fi->last_name);
220 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
222 struct ceph_file_info *fi = filp->private_data;
223 struct inode *inode = filp->f_dentry->d_inode;
224 struct ceph_inode_info *ci = ceph_inode(inode);
225 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
226 struct ceph_mds_client *mdsc = fsc->mdsc;
227 unsigned frag = fpos_frag(filp->f_pos);
228 int off = fpos_off(filp->f_pos);
231 struct ceph_mds_reply_info_parsed *rinfo;
232 const int max_entries = fsc->mount_options->max_readdir;
233 const int max_bytes = fsc->mount_options->max_readdir_bytes;
235 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
239 /* always start with . and .. */
240 if (filp->f_pos == 0) {
241 /* note dir version at start of readdir so we can tell
242 * if any dentries get dropped */
243 fi->dir_release_count = ci->i_release_count;
245 dout("readdir off 0 -> '.'\n");
246 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
247 inode->i_ino, inode->i_mode >> 12) < 0)
252 if (filp->f_pos == 1) {
253 dout("readdir off 1 -> '..'\n");
254 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
255 filp->f_dentry->d_parent->d_inode->i_ino,
256 inode->i_mode >> 12) < 0)
262 /* can we use the dcache? */
263 spin_lock(&inode->i_lock);
264 if ((filp->f_pos == 2 || fi->dentry) &&
265 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
266 ceph_snap(inode) != CEPH_SNAPDIR &&
267 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
268 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
269 spin_unlock(&inode->i_lock);
270 err = __dcache_readdir(filp, dirent, filldir);
274 spin_unlock(&inode->i_lock);
277 err = note_last_dentry(fi, fi->dentry->d_name.name,
278 fi->dentry->d_name.len);
285 /* proceed with a normal readdir */
288 /* do we have the correct frag content buffered? */
289 if (fi->frag != frag || fi->last_readdir == NULL) {
290 struct ceph_mds_request *req;
291 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
292 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
294 /* discard old result, if any */
295 if (fi->last_readdir) {
296 ceph_mdsc_put_request(fi->last_readdir);
297 fi->last_readdir = NULL;
300 /* requery frag tree, as the frag topology may have changed */
301 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
303 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
304 ceph_vinop(inode), frag, fi->last_name);
305 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
308 req->r_inode = igrab(inode);
309 req->r_dentry = dget(filp->f_dentry);
310 /* hints to request -> mds selection code */
311 req->r_direct_mode = USE_AUTH_MDS;
312 req->r_direct_hash = ceph_frag_value(frag);
313 req->r_direct_is_hash = true;
314 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
315 req->r_readdir_offset = fi->next_offset;
316 req->r_args.readdir.frag = cpu_to_le32(frag);
317 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
318 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
319 req->r_num_caps = max_entries + 1;
320 err = ceph_mdsc_do_request(mdsc, NULL, req);
322 ceph_mdsc_put_request(req);
325 dout("readdir got and parsed readdir result=%d"
326 " on frag %x, end=%d, complete=%d\n", err, frag,
327 (int)req->r_reply_info.dir_end,
328 (int)req->r_reply_info.dir_complete);
330 if (!req->r_did_prepopulate) {
331 dout("readdir !did_prepopulate");
332 fi->dir_release_count--; /* preclude I_COMPLETE */
335 /* note next offset and last dentry name */
336 fi->offset = fi->next_offset;
337 fi->last_readdir = req;
339 if (req->r_reply_info.dir_end) {
340 kfree(fi->last_name);
341 fi->last_name = NULL;
342 if (ceph_frag_is_rightmost(frag))
347 rinfo = &req->r_reply_info;
348 err = note_last_dentry(fi,
349 rinfo->dir_dname[rinfo->dir_nr-1],
350 rinfo->dir_dname_len[rinfo->dir_nr-1]);
353 fi->next_offset += rinfo->dir_nr;
357 rinfo = &fi->last_readdir->r_reply_info;
358 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
359 rinfo->dir_nr, off, fi->offset);
360 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
361 u64 pos = ceph_make_fpos(frag, off);
362 struct ceph_mds_reply_inode *in =
363 rinfo->dir_in[off - fi->offset].in;
364 struct ceph_vino vino;
367 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
368 off, off - fi->offset, rinfo->dir_nr, pos,
369 rinfo->dir_dname_len[off - fi->offset],
370 rinfo->dir_dname[off - fi->offset], in);
372 ftype = le32_to_cpu(in->mode) >> 12;
373 vino.ino = le64_to_cpu(in->ino);
374 vino.snap = le64_to_cpu(in->snapid);
375 ino = ceph_vino_to_ino(vino);
377 rinfo->dir_dname[off - fi->offset],
378 rinfo->dir_dname_len[off - fi->offset],
379 pos, ino, ftype) < 0) {
380 dout("filldir stopping us...\n");
384 filp->f_pos = pos + 1;
388 ceph_mdsc_put_request(fi->last_readdir);
389 fi->last_readdir = NULL;
394 if (!ceph_frag_is_rightmost(frag)) {
395 frag = ceph_frag_next(frag);
397 filp->f_pos = ceph_make_fpos(frag, off);
398 dout("readdir next frag is %x\n", frag);
404 * if dir_release_count still matches the dir, no dentries
405 * were released during the whole readdir, and we should have
406 * the complete dir contents in our cache.
408 spin_lock(&inode->i_lock);
409 if (ci->i_release_count == fi->dir_release_count) {
410 dout(" marking %p complete\n", inode);
411 ci->i_ceph_flags |= CEPH_I_COMPLETE;
412 ci->i_max_offset = filp->f_pos;
414 spin_unlock(&inode->i_lock);
416 dout("readdir %p filp %p done.\n", inode, filp);
420 static void reset_readdir(struct ceph_file_info *fi)
422 if (fi->last_readdir) {
423 ceph_mdsc_put_request(fi->last_readdir);
424 fi->last_readdir = NULL;
426 kfree(fi->last_name);
427 fi->last_name = NULL;
428 fi->next_offset = 2; /* compensate for . and .. */
436 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
438 struct ceph_file_info *fi = file->private_data;
439 struct inode *inode = file->f_mapping->host;
440 loff_t old_offset = offset;
443 mutex_lock(&inode->i_mutex);
446 offset += inode->i_size + 2; /* FIXME */
449 offset += file->f_pos;
452 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
453 if (offset != file->f_pos) {
454 file->f_pos = offset;
461 * discard buffered readdir content on seekdir(0), or
462 * seek to new frag, or seek prior to current chunk.
465 fpos_frag(offset) != fpos_frag(old_offset) ||
466 fpos_off(offset) < fi->offset) {
467 dout("dir_llseek dropping %p content\n", file);
471 /* bump dir_release_count if we did a forward seek */
472 if (offset > old_offset)
473 fi->dir_release_count--;
475 mutex_unlock(&inode->i_mutex);
480 * Process result of a lookup/open request.
482 * Mainly, make sure we return the final req->r_dentry (if it already
483 * existed) in place of the original VFS-provided dentry when they
486 * Gracefully handle the case where the MDS replies with -ENOENT and
487 * no trace (which it may do, at its discretion, e.g., if it doesn't
488 * care to issue a lease on the negative dentry).
490 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
491 struct dentry *dentry, int err)
493 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
494 struct inode *parent = dentry->d_parent->d_inode;
497 if (err == -ENOENT &&
498 strcmp(dentry->d_name.name,
499 fsc->mount_options->snapdir_name) == 0) {
500 struct inode *inode = ceph_get_snapdir(parent);
501 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
502 dentry, dentry->d_name.len, dentry->d_name.name, inode);
503 BUG_ON(!d_unhashed(dentry));
504 d_add(dentry, inode);
508 if (err == -ENOENT) {
511 if (!req->r_reply_info.head->is_dentry) {
512 dout("ENOENT and no trace, dentry %p inode %p\n",
513 dentry, dentry->d_inode);
514 if (dentry->d_inode) {
523 dentry = ERR_PTR(err);
524 else if (dentry != req->r_dentry)
525 dentry = dget(req->r_dentry); /* we got spliced */
531 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
533 return ceph_ino(inode) == CEPH_INO_ROOT &&
534 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
538 * Look up a single dir entry. If there is a lookup intent, inform
539 * the MDS so that it gets our 'caps wanted' value in a single op.
541 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
542 struct nameidata *nd)
544 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
545 struct ceph_mds_client *mdsc = fsc->mdsc;
546 struct ceph_mds_request *req;
550 dout("lookup %p dentry %p '%.*s'\n",
551 dir, dentry, dentry->d_name.len, dentry->d_name.name);
553 if (dentry->d_name.len > NAME_MAX)
554 return ERR_PTR(-ENAMETOOLONG);
556 err = ceph_init_dentry(dentry);
560 /* open (but not create!) intent? */
562 (nd->flags & LOOKUP_OPEN) &&
563 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
564 !(nd->intent.open.flags & O_CREAT)) {
565 int mode = nd->intent.open.create_mode & ~current->fs->umask;
566 return ceph_lookup_open(dir, dentry, nd, mode, 1);
569 /* can we conclude ENOENT locally? */
570 if (dentry->d_inode == NULL) {
571 struct ceph_inode_info *ci = ceph_inode(dir);
572 struct ceph_dentry_info *di = ceph_dentry(dentry);
574 spin_lock(&dir->i_lock);
575 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
576 if (strncmp(dentry->d_name.name,
577 fsc->mount_options->snapdir_name,
578 dentry->d_name.len) &&
579 !is_root_ceph_dentry(dir, dentry) &&
580 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
581 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
582 spin_unlock(&dir->i_lock);
583 dout(" dir %p complete, -ENOENT\n", dir);
585 di->lease_shared_gen = ci->i_shared_gen;
588 spin_unlock(&dir->i_lock);
591 op = ceph_snap(dir) == CEPH_SNAPDIR ?
592 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
593 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
595 return ERR_CAST(req);
596 req->r_dentry = dget(dentry);
598 /* we only need inode linkage */
599 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
600 req->r_locked_dir = dir;
601 err = ceph_mdsc_do_request(mdsc, NULL, req);
602 dentry = ceph_finish_lookup(req, dentry, err);
603 ceph_mdsc_put_request(req); /* will dput(dentry) */
604 dout("lookup result=%p\n", dentry);
609 * If we do a create but get no trace back from the MDS, follow up with
610 * a lookup (the VFS expects us to link up the provided dentry).
612 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
614 struct dentry *result = ceph_lookup(dir, dentry, NULL);
616 if (result && !IS_ERR(result)) {
618 * We created the item, then did a lookup, and found
619 * it was already linked to another inode we already
620 * had in our cache (and thus got spliced). Link our
621 * dentry to that inode, but don't hash it, just in
622 * case the VFS wants to dereference it.
624 BUG_ON(!result->d_inode);
625 d_instantiate(dentry, result->d_inode);
628 return PTR_ERR(result);
631 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
632 int mode, dev_t rdev)
634 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
635 struct ceph_mds_client *mdsc = fsc->mdsc;
636 struct ceph_mds_request *req;
639 if (ceph_snap(dir) != CEPH_NOSNAP)
642 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
643 dir, dentry, mode, rdev);
644 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
649 req->r_dentry = dget(dentry);
651 req->r_locked_dir = dir;
652 req->r_args.mknod.mode = cpu_to_le32(mode);
653 req->r_args.mknod.rdev = cpu_to_le32(rdev);
654 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
655 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
656 err = ceph_mdsc_do_request(mdsc, dir, req);
657 if (!err && !req->r_reply_info.head->is_dentry)
658 err = ceph_handle_notrace_create(dir, dentry);
659 ceph_mdsc_put_request(req);
665 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
666 struct nameidata *nd)
668 dout("create in dir %p dentry %p name '%.*s'\n",
669 dir, dentry, dentry->d_name.len, dentry->d_name.name);
671 if (ceph_snap(dir) != CEPH_NOSNAP)
675 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
676 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
677 /* hrm, what should i do here if we get aliased? */
679 return PTR_ERR(dentry);
683 /* fall back to mknod */
684 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
687 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
690 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
691 struct ceph_mds_client *mdsc = fsc->mdsc;
692 struct ceph_mds_request *req;
695 if (ceph_snap(dir) != CEPH_NOSNAP)
698 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
699 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
704 req->r_dentry = dget(dentry);
706 req->r_path2 = kstrdup(dest, GFP_NOFS);
707 req->r_locked_dir = dir;
708 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
709 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
710 err = ceph_mdsc_do_request(mdsc, dir, req);
711 if (!err && !req->r_reply_info.head->is_dentry)
712 err = ceph_handle_notrace_create(dir, dentry);
713 ceph_mdsc_put_request(req);
719 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
721 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
722 struct ceph_mds_client *mdsc = fsc->mdsc;
723 struct ceph_mds_request *req;
727 if (ceph_snap(dir) == CEPH_SNAPDIR) {
728 /* mkdir .snap/foo is a MKSNAP */
729 op = CEPH_MDS_OP_MKSNAP;
730 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
731 dentry->d_name.len, dentry->d_name.name, dentry);
732 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
733 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
734 op = CEPH_MDS_OP_MKDIR;
738 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
744 req->r_dentry = dget(dentry);
746 req->r_locked_dir = dir;
747 req->r_args.mkdir.mode = cpu_to_le32(mode);
748 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
749 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
750 err = ceph_mdsc_do_request(mdsc, dir, req);
751 if (!err && !req->r_reply_info.head->is_dentry)
752 err = ceph_handle_notrace_create(dir, dentry);
753 ceph_mdsc_put_request(req);
760 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
761 struct dentry *dentry)
763 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
764 struct ceph_mds_client *mdsc = fsc->mdsc;
765 struct ceph_mds_request *req;
768 if (ceph_snap(dir) != CEPH_NOSNAP)
771 dout("link in dir %p old_dentry %p dentry %p\n", dir,
773 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
778 req->r_dentry = dget(dentry);
780 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
781 req->r_locked_dir = dir;
782 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
783 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
784 err = ceph_mdsc_do_request(mdsc, dir, req);
787 else if (!req->r_reply_info.head->is_dentry)
788 d_instantiate(dentry, igrab(old_dentry->d_inode));
789 ceph_mdsc_put_request(req);
794 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
795 * looks like the link count will hit 0, drop any other caps (other
796 * than PIN) we don't specifically want (due to the file still being
799 static int drop_caps_for_unlink(struct inode *inode)
801 struct ceph_inode_info *ci = ceph_inode(inode);
802 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
804 spin_lock(&inode->i_lock);
805 if (inode->i_nlink == 1) {
806 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
807 ci->i_ceph_flags |= CEPH_I_NODELAY;
809 spin_unlock(&inode->i_lock);
814 * rmdir and unlink are differ only by the metadata op code
816 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
818 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
819 struct ceph_mds_client *mdsc = fsc->mdsc;
820 struct inode *inode = dentry->d_inode;
821 struct ceph_mds_request *req;
825 if (ceph_snap(dir) == CEPH_SNAPDIR) {
826 /* rmdir .snap/foo is RMSNAP */
827 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
828 dentry->d_name.name, dentry);
829 op = CEPH_MDS_OP_RMSNAP;
830 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
831 dout("unlink/rmdir dir %p dn %p inode %p\n",
833 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
834 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
837 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
842 req->r_dentry = dget(dentry);
844 req->r_locked_dir = dir;
845 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
846 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
847 req->r_inode_drop = drop_caps_for_unlink(inode);
848 err = ceph_mdsc_do_request(mdsc, dir, req);
849 if (!err && !req->r_reply_info.head->is_dentry)
851 ceph_mdsc_put_request(req);
856 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
857 struct inode *new_dir, struct dentry *new_dentry)
859 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
860 struct ceph_mds_client *mdsc = fsc->mdsc;
861 struct ceph_mds_request *req;
864 if (ceph_snap(old_dir) != ceph_snap(new_dir))
866 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
867 ceph_snap(new_dir) != CEPH_NOSNAP)
869 dout("rename dir %p dentry %p to dir %p dentry %p\n",
870 old_dir, old_dentry, new_dir, new_dentry);
871 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
874 req->r_dentry = dget(new_dentry);
876 req->r_old_dentry = dget(old_dentry);
877 req->r_locked_dir = new_dir;
878 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
879 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
880 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
881 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
882 /* release LINK_RDCACHE on source inode (mds will lock it) */
883 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
884 if (new_dentry->d_inode)
885 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
886 err = ceph_mdsc_do_request(mdsc, old_dir, req);
887 if (!err && !req->r_reply_info.head->is_dentry) {
889 * Normally d_move() is done by fill_trace (called by
890 * do_request, above). If there is no trace, we need
894 /* d_move screws up d_subdirs order */
895 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
897 d_move(old_dentry, new_dentry);
899 /* ensure target dentry is invalidated, despite
900 rehashing bug in vfs_rename_dir */
901 ceph_invalidate_dentry_lease(new_dentry);
903 ceph_mdsc_put_request(req);
908 * Ensure a dentry lease will no longer revalidate.
910 void ceph_invalidate_dentry_lease(struct dentry *dentry)
912 spin_lock(&dentry->d_lock);
913 dentry->d_time = jiffies;
914 ceph_dentry(dentry)->lease_shared_gen = 0;
915 spin_unlock(&dentry->d_lock);
919 * Check if dentry lease is valid. If not, delete the lease. Try to
920 * renew if the least is more than half up.
922 static int dentry_lease_is_valid(struct dentry *dentry)
924 struct ceph_dentry_info *di;
925 struct ceph_mds_session *s;
929 struct ceph_mds_session *session = NULL;
930 struct inode *dir = NULL;
933 spin_lock(&dentry->d_lock);
934 di = ceph_dentry(dentry);
935 if (di && di->lease_session) {
936 s = di->lease_session;
937 spin_lock(&s->s_cap_lock);
940 spin_unlock(&s->s_cap_lock);
942 if (di->lease_gen == gen &&
943 time_before(jiffies, dentry->d_time) &&
944 time_before(jiffies, ttl)) {
946 if (di->lease_renew_after &&
947 time_after(jiffies, di->lease_renew_after)) {
948 /* we should renew */
949 dir = dentry->d_parent->d_inode;
950 session = ceph_get_mds_session(s);
952 di->lease_renew_after = 0;
953 di->lease_renew_from = jiffies;
957 spin_unlock(&dentry->d_lock);
960 ceph_mdsc_lease_send_msg(session, dir, dentry,
961 CEPH_MDS_LEASE_RENEW, seq);
962 ceph_put_mds_session(session);
964 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
969 * Check if directory-wide content lease/cap is valid.
971 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
973 struct ceph_inode_info *ci = ceph_inode(dir);
974 struct ceph_dentry_info *di = ceph_dentry(dentry);
977 spin_lock(&dir->i_lock);
978 if (ci->i_shared_gen == di->lease_shared_gen)
979 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
980 spin_unlock(&dir->i_lock);
981 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
982 dir, (unsigned)ci->i_shared_gen, dentry,
983 (unsigned)di->lease_shared_gen, valid);
988 * Check if cached dentry can be trusted.
990 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
992 struct inode *dir = dentry->d_parent->d_inode;
994 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
995 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
996 ceph_dentry(dentry)->offset);
998 /* always trust cached snapped dentries, snapdir dentry */
999 if (ceph_snap(dir) != CEPH_NOSNAP) {
1000 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1001 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1004 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1007 if (dentry_lease_is_valid(dentry) ||
1008 dir_lease_is_valid(dir, dentry))
1011 dout("d_revalidate %p invalid\n", dentry);
1015 ceph_dentry_lru_touch(dentry);
1020 * When a dentry is released, clear the dir I_COMPLETE if it was part
1021 * of the current dir gen or if this is in the snapshot namespace.
1023 static void ceph_dentry_release(struct dentry *dentry)
1025 struct ceph_dentry_info *di = ceph_dentry(dentry);
1026 struct inode *parent_inode = NULL;
1027 u64 snapid = CEPH_NOSNAP;
1029 if (!IS_ROOT(dentry)) {
1030 parent_inode = dentry->d_parent->d_inode;
1032 snapid = ceph_snap(parent_inode);
1034 dout("dentry_release %p parent %p\n", dentry, parent_inode);
1035 if (parent_inode && snapid != CEPH_SNAPDIR) {
1036 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1038 spin_lock(&parent_inode->i_lock);
1039 if (ci->i_shared_gen == di->lease_shared_gen ||
1040 snapid <= CEPH_MAXSNAP) {
1041 dout(" clearing %p complete (d_release)\n",
1043 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1044 ci->i_release_count++;
1046 spin_unlock(&parent_inode->i_lock);
1049 ceph_dentry_lru_del(dentry);
1050 if (di->lease_session)
1051 ceph_put_mds_session(di->lease_session);
1052 kmem_cache_free(ceph_dentry_cachep, di);
1053 dentry->d_fsdata = NULL;
1057 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1058 struct nameidata *nd)
1061 * Eventually, we'll want to revalidate snapped metadata
1062 * too... probably...
1070 * read() on a dir. This weird interface hack only works if mounted
1071 * with '-o dirstat'.
1073 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1076 struct ceph_file_info *cf = file->private_data;
1077 struct inode *inode = file->f_dentry->d_inode;
1078 struct ceph_inode_info *ci = ceph_inode(inode);
1081 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1084 if (!cf->dir_info) {
1085 cf->dir_info = kmalloc(1024, GFP_NOFS);
1089 sprintf(cf->dir_info,
1092 " subdirs: %20lld\n"
1093 "rentries: %20lld\n"
1095 " rsubdirs: %20lld\n"
1097 "rctime: %10ld.%09ld\n",
1098 ci->i_files + ci->i_subdirs,
1101 ci->i_rfiles + ci->i_rsubdirs,
1105 (long)ci->i_rctime.tv_sec,
1106 (long)ci->i_rctime.tv_nsec);
1109 if (*ppos >= cf->dir_info_len)
1111 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1112 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1115 *ppos += (size - left);
1120 * an fsync() on a dir will wait for any uncommitted directory
1121 * operations to commit.
1123 static int ceph_dir_fsync(struct file *file, int datasync)
1125 struct inode *inode = file->f_path.dentry->d_inode;
1126 struct ceph_inode_info *ci = ceph_inode(inode);
1127 struct list_head *head = &ci->i_unsafe_dirops;
1128 struct ceph_mds_request *req;
1132 dout("dir_fsync %p\n", inode);
1133 spin_lock(&ci->i_unsafe_lock);
1134 if (list_empty(head))
1137 req = list_entry(head->prev,
1138 struct ceph_mds_request, r_unsafe_dir_item);
1139 last_tid = req->r_tid;
1142 ceph_mdsc_get_request(req);
1143 spin_unlock(&ci->i_unsafe_lock);
1144 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1145 inode, req->r_tid, last_tid);
1146 if (req->r_timeout) {
1147 ret = wait_for_completion_timeout(
1148 &req->r_safe_completion, req->r_timeout);
1152 ret = -EIO; /* timed out */
1154 wait_for_completion(&req->r_safe_completion);
1156 spin_lock(&ci->i_unsafe_lock);
1157 ceph_mdsc_put_request(req);
1159 if (ret || list_empty(head))
1161 req = list_entry(head->next,
1162 struct ceph_mds_request, r_unsafe_dir_item);
1163 } while (req->r_tid < last_tid);
1165 spin_unlock(&ci->i_unsafe_lock);
1170 * We maintain a private dentry LRU.
1172 * FIXME: this needs to be changed to a per-mds lru to be useful.
1174 void ceph_dentry_lru_add(struct dentry *dn)
1176 struct ceph_dentry_info *di = ceph_dentry(dn);
1177 struct ceph_mds_client *mdsc;
1179 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1180 dn->d_name.len, dn->d_name.name);
1182 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1183 spin_lock(&mdsc->dentry_lru_lock);
1184 list_add_tail(&di->lru, &mdsc->dentry_lru);
1186 spin_unlock(&mdsc->dentry_lru_lock);
1190 void ceph_dentry_lru_touch(struct dentry *dn)
1192 struct ceph_dentry_info *di = ceph_dentry(dn);
1193 struct ceph_mds_client *mdsc;
1195 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1196 dn->d_name.len, dn->d_name.name, di->offset);
1198 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1199 spin_lock(&mdsc->dentry_lru_lock);
1200 list_move_tail(&di->lru, &mdsc->dentry_lru);
1201 spin_unlock(&mdsc->dentry_lru_lock);
1205 void ceph_dentry_lru_del(struct dentry *dn)
1207 struct ceph_dentry_info *di = ceph_dentry(dn);
1208 struct ceph_mds_client *mdsc;
1210 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1211 dn->d_name.len, dn->d_name.name);
1213 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1214 spin_lock(&mdsc->dentry_lru_lock);
1215 list_del_init(&di->lru);
1217 spin_unlock(&mdsc->dentry_lru_lock);
1221 const struct file_operations ceph_dir_fops = {
1222 .read = ceph_read_dir,
1223 .readdir = ceph_readdir,
1224 .llseek = ceph_dir_llseek,
1226 .release = ceph_release,
1227 .unlocked_ioctl = ceph_ioctl,
1228 .fsync = ceph_dir_fsync,
1231 const struct inode_operations ceph_dir_iops = {
1232 .lookup = ceph_lookup,
1233 .permission = ceph_permission,
1234 .getattr = ceph_getattr,
1235 .setattr = ceph_setattr,
1236 .setxattr = ceph_setxattr,
1237 .getxattr = ceph_getxattr,
1238 .listxattr = ceph_listxattr,
1239 .removexattr = ceph_removexattr,
1240 .mknod = ceph_mknod,
1241 .symlink = ceph_symlink,
1242 .mkdir = ceph_mkdir,
1244 .unlink = ceph_unlink,
1245 .rmdir = ceph_unlink,
1246 .rename = ceph_rename,
1247 .create = ceph_create,
1250 const struct dentry_operations ceph_dentry_ops = {
1251 .d_revalidate = ceph_d_revalidate,
1252 .d_release = ceph_dentry_release,
1255 const struct dentry_operations ceph_snapdir_dentry_ops = {
1256 .d_revalidate = ceph_snapdir_d_revalidate,
1257 .d_release = ceph_dentry_release,
1260 const struct dentry_operations ceph_snap_dentry_ops = {
1261 .d_release = ceph_dentry_release,