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;
104 * When possible, we try to satisfy a readdir by peeking at the
105 * dcache. We make this work by carefully ordering dentries on
106 * d_u.d_child when we initially get results back from the MDS, and
107 * falling back to a "normal" sync readdir if any dentries in the dir
110 * I_COMPLETE tells indicates we have all dentries in the dir. It is
111 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
112 * the MDS if/when the directory is modified).
114 static int __dcache_readdir(struct file *filp,
115 void *dirent, filldir_t filldir)
117 struct ceph_file_info *fi = filp->private_data;
118 struct dentry *parent = filp->f_dentry;
119 struct inode *dir = parent->d_inode;
121 struct dentry *dentry, *last;
122 struct ceph_dentry_info *di;
125 /* claim ref on last dentry we returned */
129 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
132 spin_lock(&parent->d_lock);
134 /* start at beginning? */
135 if (filp->f_pos == 2 || last == NULL ||
136 filp->f_pos < ceph_dentry(last)->offset) {
137 if (list_empty(&parent->d_subdirs))
139 p = parent->d_subdirs.prev;
140 dout(" initial p %p/%p\n", p->prev, p->next);
142 p = last->d_u.d_child.prev;
146 dentry = list_entry(p, struct dentry, d_u.d_child);
147 di = ceph_dentry(dentry);
149 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
150 d_unhashed(dentry) ? "!hashed" : "hashed",
151 parent->d_subdirs.prev, parent->d_subdirs.next);
152 if (p == &parent->d_subdirs) {
153 fi->flags |= CEPH_F_ATEND;
156 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
157 if (!d_unhashed(dentry) && dentry->d_inode &&
158 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
159 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
160 filp->f_pos <= di->offset)
162 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
163 dentry->d_name.len, dentry->d_name.name, di->offset,
164 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
165 !dentry->d_inode ? " null" : "");
166 spin_unlock(&dentry->d_lock);
168 dentry = list_entry(p, struct dentry, d_u.d_child);
169 di = ceph_dentry(dentry);
173 spin_unlock(&dentry->d_lock);
174 spin_unlock(&parent->d_lock);
176 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
177 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
178 filp->f_pos = di->offset;
179 err = filldir(dirent, dentry->d_name.name,
180 dentry->d_name.len, di->offset,
181 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
182 dentry->d_inode->i_mode >> 12);
186 /* remember our position */
188 fi->next_offset = di->offset;
200 /* make sure a dentry wasn't dropped while we didn't have parent lock */
201 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
202 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
207 spin_lock(&parent->d_lock);
208 p = p->prev; /* advance to next dentry */
212 spin_unlock(&parent->d_lock);
220 * make note of the last dentry we read, so we can
221 * continue at the same lexicographical point,
222 * regardless of what dir changes take place on the
225 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
228 kfree(fi->last_name);
229 fi->last_name = kmalloc(len+1, GFP_NOFS);
232 memcpy(fi->last_name, name, len);
233 fi->last_name[len] = 0;
234 dout("note_last_dentry '%s'\n", fi->last_name);
238 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
240 struct ceph_file_info *fi = filp->private_data;
241 struct inode *inode = file_inode(filp);
242 struct ceph_inode_info *ci = ceph_inode(inode);
243 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
244 struct ceph_mds_client *mdsc = fsc->mdsc;
245 unsigned frag = fpos_frag(filp->f_pos);
246 int off = fpos_off(filp->f_pos);
249 struct ceph_mds_reply_info_parsed *rinfo;
250 const int max_entries = fsc->mount_options->max_readdir;
251 const int max_bytes = fsc->mount_options->max_readdir_bytes;
253 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
254 if (fi->flags & CEPH_F_ATEND)
257 /* always start with . and .. */
258 if (filp->f_pos == 0) {
259 /* note dir version at start of readdir so we can tell
260 * if any dentries get dropped */
261 fi->dir_release_count = ci->i_release_count;
263 dout("readdir off 0 -> '.'\n");
264 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
265 ceph_translate_ino(inode->i_sb, inode->i_ino),
266 inode->i_mode >> 12) < 0)
271 if (filp->f_pos == 1) {
272 ino_t ino = parent_ino(filp->f_dentry);
273 dout("readdir off 1 -> '..'\n");
274 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
275 ceph_translate_ino(inode->i_sb, ino),
276 inode->i_mode >> 12) < 0)
282 /* can we use the dcache? */
283 spin_lock(&ci->i_ceph_lock);
284 if ((filp->f_pos == 2 || fi->dentry) &&
285 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
286 ceph_snap(inode) != CEPH_SNAPDIR &&
287 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
288 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
289 spin_unlock(&ci->i_ceph_lock);
290 err = __dcache_readdir(filp, dirent, filldir);
294 spin_unlock(&ci->i_ceph_lock);
297 err = note_last_dentry(fi, fi->dentry->d_name.name,
298 fi->dentry->d_name.len);
305 /* proceed with a normal readdir */
308 /* do we have the correct frag content buffered? */
309 if (fi->frag != frag || fi->last_readdir == NULL) {
310 struct ceph_mds_request *req;
311 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
312 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
314 /* discard old result, if any */
315 if (fi->last_readdir) {
316 ceph_mdsc_put_request(fi->last_readdir);
317 fi->last_readdir = NULL;
320 /* requery frag tree, as the frag topology may have changed */
321 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
323 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
324 ceph_vinop(inode), frag, fi->last_name);
325 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
328 req->r_inode = inode;
330 req->r_dentry = dget(filp->f_dentry);
331 /* hints to request -> mds selection code */
332 req->r_direct_mode = USE_AUTH_MDS;
333 req->r_direct_hash = ceph_frag_value(frag);
334 req->r_direct_is_hash = true;
335 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
336 req->r_readdir_offset = fi->next_offset;
337 req->r_args.readdir.frag = cpu_to_le32(frag);
338 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
339 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
340 req->r_num_caps = max_entries + 1;
341 err = ceph_mdsc_do_request(mdsc, NULL, req);
343 ceph_mdsc_put_request(req);
346 dout("readdir got and parsed readdir result=%d"
347 " on frag %x, end=%d, complete=%d\n", err, frag,
348 (int)req->r_reply_info.dir_end,
349 (int)req->r_reply_info.dir_complete);
351 if (!req->r_did_prepopulate) {
352 dout("readdir !did_prepopulate");
353 fi->dir_release_count--; /* preclude I_COMPLETE */
356 /* note next offset and last dentry name */
357 fi->offset = fi->next_offset;
358 fi->last_readdir = req;
360 if (req->r_reply_info.dir_end) {
361 kfree(fi->last_name);
362 fi->last_name = NULL;
363 if (ceph_frag_is_rightmost(frag))
368 rinfo = &req->r_reply_info;
369 err = note_last_dentry(fi,
370 rinfo->dir_dname[rinfo->dir_nr-1],
371 rinfo->dir_dname_len[rinfo->dir_nr-1]);
374 fi->next_offset += rinfo->dir_nr;
378 rinfo = &fi->last_readdir->r_reply_info;
379 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
380 rinfo->dir_nr, off, fi->offset);
381 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
382 u64 pos = ceph_make_fpos(frag, off);
383 struct ceph_mds_reply_inode *in =
384 rinfo->dir_in[off - fi->offset].in;
385 struct ceph_vino vino;
388 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
389 off, off - fi->offset, rinfo->dir_nr, pos,
390 rinfo->dir_dname_len[off - fi->offset],
391 rinfo->dir_dname[off - fi->offset], in);
393 ftype = le32_to_cpu(in->mode) >> 12;
394 vino.ino = le64_to_cpu(in->ino);
395 vino.snap = le64_to_cpu(in->snapid);
396 ino = ceph_vino_to_ino(vino);
398 rinfo->dir_dname[off - fi->offset],
399 rinfo->dir_dname_len[off - fi->offset],
401 ceph_translate_ino(inode->i_sb, ino), ftype) < 0) {
402 dout("filldir stopping us...\n");
406 filp->f_pos = pos + 1;
410 ceph_mdsc_put_request(fi->last_readdir);
411 fi->last_readdir = NULL;
416 if (!ceph_frag_is_rightmost(frag)) {
417 frag = ceph_frag_next(frag);
419 filp->f_pos = ceph_make_fpos(frag, off);
420 dout("readdir next frag is %x\n", frag);
423 fi->flags |= CEPH_F_ATEND;
426 * if dir_release_count still matches the dir, no dentries
427 * were released during the whole readdir, and we should have
428 * the complete dir contents in our cache.
430 spin_lock(&ci->i_ceph_lock);
431 if (ci->i_release_count == fi->dir_release_count) {
432 dout(" marking %p complete\n", inode);
433 ci->i_ceph_flags |= CEPH_I_COMPLETE;
434 ci->i_max_offset = filp->f_pos;
436 spin_unlock(&ci->i_ceph_lock);
438 dout("readdir %p filp %p done.\n", inode, filp);
442 static void reset_readdir(struct ceph_file_info *fi)
444 if (fi->last_readdir) {
445 ceph_mdsc_put_request(fi->last_readdir);
446 fi->last_readdir = NULL;
448 kfree(fi->last_name);
449 fi->last_name = NULL;
450 fi->next_offset = 2; /* compensate for . and .. */
455 fi->flags &= ~CEPH_F_ATEND;
458 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
460 struct ceph_file_info *fi = file->private_data;
461 struct inode *inode = file->f_mapping->host;
462 loff_t old_offset = offset;
465 mutex_lock(&inode->i_mutex);
469 offset += inode->i_size + 2; /* FIXME */
472 offset += file->f_pos;
479 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
480 if (offset != file->f_pos) {
481 file->f_pos = offset;
483 fi->flags &= ~CEPH_F_ATEND;
488 * discard buffered readdir content on seekdir(0), or
489 * seek to new frag, or seek prior to current chunk.
492 fpos_frag(offset) != fpos_frag(old_offset) ||
493 fpos_off(offset) < fi->offset) {
494 dout("dir_llseek dropping %p content\n", file);
498 /* bump dir_release_count if we did a forward seek */
499 if (offset > old_offset)
500 fi->dir_release_count--;
503 mutex_unlock(&inode->i_mutex);
508 * Handle lookups for the hidden .snap directory.
510 int ceph_handle_snapdir(struct ceph_mds_request *req,
511 struct dentry *dentry, int err)
513 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
514 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
517 if (err == -ENOENT &&
518 ceph_snap(parent) == CEPH_NOSNAP &&
519 strcmp(dentry->d_name.name,
520 fsc->mount_options->snapdir_name) == 0) {
521 struct inode *inode = ceph_get_snapdir(parent);
522 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
523 dentry, dentry->d_name.len, dentry->d_name.name, inode);
524 BUG_ON(!d_unhashed(dentry));
525 d_add(dentry, inode);
532 * Figure out final result of a lookup/open request.
534 * Mainly, make sure we return the final req->r_dentry (if it already
535 * existed) in place of the original VFS-provided dentry when they
538 * Gracefully handle the case where the MDS replies with -ENOENT and
539 * no trace (which it may do, at its discretion, e.g., if it doesn't
540 * care to issue a lease on the negative dentry).
542 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
543 struct dentry *dentry, int err)
545 if (err == -ENOENT) {
548 if (!req->r_reply_info.head->is_dentry) {
549 dout("ENOENT and no trace, dentry %p inode %p\n",
550 dentry, dentry->d_inode);
551 if (dentry->d_inode) {
560 dentry = ERR_PTR(err);
561 else if (dentry != req->r_dentry)
562 dentry = dget(req->r_dentry); /* we got spliced */
568 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
570 return ceph_ino(inode) == CEPH_INO_ROOT &&
571 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
575 * Look up a single dir entry. If there is a lookup intent, inform
576 * the MDS so that it gets our 'caps wanted' value in a single op.
578 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
581 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
582 struct ceph_mds_client *mdsc = fsc->mdsc;
583 struct ceph_mds_request *req;
587 dout("lookup %p dentry %p '%.*s'\n",
588 dir, dentry, dentry->d_name.len, dentry->d_name.name);
590 if (dentry->d_name.len > NAME_MAX)
591 return ERR_PTR(-ENAMETOOLONG);
593 err = ceph_init_dentry(dentry);
597 /* can we conclude ENOENT locally? */
598 if (dentry->d_inode == NULL) {
599 struct ceph_inode_info *ci = ceph_inode(dir);
600 struct ceph_dentry_info *di = ceph_dentry(dentry);
602 spin_lock(&ci->i_ceph_lock);
603 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
604 if (strncmp(dentry->d_name.name,
605 fsc->mount_options->snapdir_name,
606 dentry->d_name.len) &&
607 !is_root_ceph_dentry(dir, dentry) &&
608 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
609 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
610 spin_unlock(&ci->i_ceph_lock);
611 dout(" dir %p complete, -ENOENT\n", dir);
613 di->lease_shared_gen = ci->i_shared_gen;
616 spin_unlock(&ci->i_ceph_lock);
619 op = ceph_snap(dir) == CEPH_SNAPDIR ?
620 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
621 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
623 return ERR_CAST(req);
624 req->r_dentry = dget(dentry);
626 /* we only need inode linkage */
627 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
628 req->r_locked_dir = dir;
629 err = ceph_mdsc_do_request(mdsc, NULL, req);
630 err = ceph_handle_snapdir(req, dentry, err);
631 dentry = ceph_finish_lookup(req, dentry, err);
632 ceph_mdsc_put_request(req); /* will dput(dentry) */
633 dout("lookup result=%p\n", dentry);
638 * If we do a create but get no trace back from the MDS, follow up with
639 * a lookup (the VFS expects us to link up the provided dentry).
641 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
643 struct dentry *result = ceph_lookup(dir, dentry, 0);
645 if (result && !IS_ERR(result)) {
647 * We created the item, then did a lookup, and found
648 * it was already linked to another inode we already
649 * had in our cache (and thus got spliced). Link our
650 * dentry to that inode, but don't hash it, just in
651 * case the VFS wants to dereference it.
653 BUG_ON(!result->d_inode);
654 d_instantiate(dentry, result->d_inode);
657 return PTR_ERR(result);
660 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
661 umode_t mode, dev_t rdev)
663 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
664 struct ceph_mds_client *mdsc = fsc->mdsc;
665 struct ceph_mds_request *req;
668 if (ceph_snap(dir) != CEPH_NOSNAP)
671 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
672 dir, dentry, mode, rdev);
673 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
678 req->r_dentry = dget(dentry);
680 req->r_locked_dir = dir;
681 req->r_args.mknod.mode = cpu_to_le32(mode);
682 req->r_args.mknod.rdev = cpu_to_le32(rdev);
683 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
684 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
685 err = ceph_mdsc_do_request(mdsc, dir, req);
686 if (!err && !req->r_reply_info.head->is_dentry)
687 err = ceph_handle_notrace_create(dir, dentry);
688 ceph_mdsc_put_request(req);
694 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
697 return ceph_mknod(dir, dentry, mode, 0);
700 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
703 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
704 struct ceph_mds_client *mdsc = fsc->mdsc;
705 struct ceph_mds_request *req;
708 if (ceph_snap(dir) != CEPH_NOSNAP)
711 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
712 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
717 req->r_dentry = dget(dentry);
719 req->r_path2 = kstrdup(dest, GFP_NOFS);
720 req->r_locked_dir = dir;
721 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
722 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
723 err = ceph_mdsc_do_request(mdsc, dir, req);
724 if (!err && !req->r_reply_info.head->is_dentry)
725 err = ceph_handle_notrace_create(dir, dentry);
726 ceph_mdsc_put_request(req);
732 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
734 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
735 struct ceph_mds_client *mdsc = fsc->mdsc;
736 struct ceph_mds_request *req;
740 if (ceph_snap(dir) == CEPH_SNAPDIR) {
741 /* mkdir .snap/foo is a MKSNAP */
742 op = CEPH_MDS_OP_MKSNAP;
743 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
744 dentry->d_name.len, dentry->d_name.name, dentry);
745 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
746 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
747 op = CEPH_MDS_OP_MKDIR;
751 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
757 req->r_dentry = dget(dentry);
759 req->r_locked_dir = dir;
760 req->r_args.mkdir.mode = cpu_to_le32(mode);
761 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
762 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
763 err = ceph_mdsc_do_request(mdsc, dir, req);
764 if (!err && !req->r_reply_info.head->is_dentry)
765 err = ceph_handle_notrace_create(dir, dentry);
766 ceph_mdsc_put_request(req);
773 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
774 struct dentry *dentry)
776 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
777 struct ceph_mds_client *mdsc = fsc->mdsc;
778 struct ceph_mds_request *req;
781 if (ceph_snap(dir) != CEPH_NOSNAP)
784 dout("link in dir %p old_dentry %p dentry %p\n", dir,
786 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
791 req->r_dentry = dget(dentry);
793 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
794 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
795 req->r_locked_dir = dir;
796 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
797 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
798 err = ceph_mdsc_do_request(mdsc, dir, req);
801 } else if (!req->r_reply_info.head->is_dentry) {
802 ihold(old_dentry->d_inode);
803 d_instantiate(dentry, old_dentry->d_inode);
805 ceph_mdsc_put_request(req);
810 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
811 * looks like the link count will hit 0, drop any other caps (other
812 * than PIN) we don't specifically want (due to the file still being
815 static int drop_caps_for_unlink(struct inode *inode)
817 struct ceph_inode_info *ci = ceph_inode(inode);
818 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
820 spin_lock(&ci->i_ceph_lock);
821 if (inode->i_nlink == 1) {
822 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
823 ci->i_ceph_flags |= CEPH_I_NODELAY;
825 spin_unlock(&ci->i_ceph_lock);
830 * rmdir and unlink are differ only by the metadata op code
832 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
834 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
835 struct ceph_mds_client *mdsc = fsc->mdsc;
836 struct inode *inode = dentry->d_inode;
837 struct ceph_mds_request *req;
841 if (ceph_snap(dir) == CEPH_SNAPDIR) {
842 /* rmdir .snap/foo is RMSNAP */
843 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
844 dentry->d_name.name, dentry);
845 op = CEPH_MDS_OP_RMSNAP;
846 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
847 dout("unlink/rmdir dir %p dn %p inode %p\n",
849 op = S_ISDIR(dentry->d_inode->i_mode) ?
850 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
853 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
858 req->r_dentry = dget(dentry);
860 req->r_locked_dir = dir;
861 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
862 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
863 req->r_inode_drop = drop_caps_for_unlink(inode);
864 err = ceph_mdsc_do_request(mdsc, dir, req);
865 if (!err && !req->r_reply_info.head->is_dentry)
867 ceph_mdsc_put_request(req);
872 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
873 struct inode *new_dir, struct dentry *new_dentry)
875 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
876 struct ceph_mds_client *mdsc = fsc->mdsc;
877 struct ceph_mds_request *req;
880 if (ceph_snap(old_dir) != ceph_snap(new_dir))
882 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
883 ceph_snap(new_dir) != CEPH_NOSNAP)
885 dout("rename dir %p dentry %p to dir %p dentry %p\n",
886 old_dir, old_dentry, new_dir, new_dentry);
887 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
890 req->r_dentry = dget(new_dentry);
892 req->r_old_dentry = dget(old_dentry);
893 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
894 req->r_locked_dir = new_dir;
895 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
896 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
897 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
898 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
899 /* release LINK_RDCACHE on source inode (mds will lock it) */
900 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
901 if (new_dentry->d_inode)
902 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
903 err = ceph_mdsc_do_request(mdsc, old_dir, req);
904 if (!err && !req->r_reply_info.head->is_dentry) {
906 * Normally d_move() is done by fill_trace (called by
907 * do_request, above). If there is no trace, we need
911 /* d_move screws up d_subdirs order */
912 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
914 d_move(old_dentry, new_dentry);
916 /* ensure target dentry is invalidated, despite
917 rehashing bug in vfs_rename_dir */
918 ceph_invalidate_dentry_lease(new_dentry);
920 ceph_mdsc_put_request(req);
925 * Ensure a dentry lease will no longer revalidate.
927 void ceph_invalidate_dentry_lease(struct dentry *dentry)
929 spin_lock(&dentry->d_lock);
930 dentry->d_time = jiffies;
931 ceph_dentry(dentry)->lease_shared_gen = 0;
932 spin_unlock(&dentry->d_lock);
936 * Check if dentry lease is valid. If not, delete the lease. Try to
937 * renew if the least is more than half up.
939 static int dentry_lease_is_valid(struct dentry *dentry)
941 struct ceph_dentry_info *di;
942 struct ceph_mds_session *s;
946 struct ceph_mds_session *session = NULL;
947 struct inode *dir = NULL;
950 spin_lock(&dentry->d_lock);
951 di = ceph_dentry(dentry);
952 if (di->lease_session) {
953 s = di->lease_session;
954 spin_lock(&s->s_gen_ttl_lock);
957 spin_unlock(&s->s_gen_ttl_lock);
959 if (di->lease_gen == gen &&
960 time_before(jiffies, dentry->d_time) &&
961 time_before(jiffies, ttl)) {
963 if (di->lease_renew_after &&
964 time_after(jiffies, di->lease_renew_after)) {
965 /* we should renew */
966 dir = dentry->d_parent->d_inode;
967 session = ceph_get_mds_session(s);
969 di->lease_renew_after = 0;
970 di->lease_renew_from = jiffies;
974 spin_unlock(&dentry->d_lock);
977 ceph_mdsc_lease_send_msg(session, dir, dentry,
978 CEPH_MDS_LEASE_RENEW, seq);
979 ceph_put_mds_session(session);
981 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
986 * Check if directory-wide content lease/cap is valid.
988 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
990 struct ceph_inode_info *ci = ceph_inode(dir);
991 struct ceph_dentry_info *di = ceph_dentry(dentry);
994 spin_lock(&ci->i_ceph_lock);
995 if (ci->i_shared_gen == di->lease_shared_gen)
996 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
997 spin_unlock(&ci->i_ceph_lock);
998 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
999 dir, (unsigned)ci->i_shared_gen, dentry,
1000 (unsigned)di->lease_shared_gen, valid);
1005 * Check if cached dentry can be trusted.
1007 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1012 if (flags & LOOKUP_RCU)
1015 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1016 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1017 ceph_dentry(dentry)->offset);
1019 dir = ceph_get_dentry_parent_inode(dentry);
1021 /* always trust cached snapped dentries, snapdir dentry */
1022 if (ceph_snap(dir) != CEPH_NOSNAP) {
1023 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1024 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1026 } else if (dentry->d_inode &&
1027 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1029 } else if (dentry_lease_is_valid(dentry) ||
1030 dir_lease_is_valid(dir, dentry)) {
1034 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1036 ceph_dentry_lru_touch(dentry);
1044 * Release our ceph_dentry_info.
1046 static void ceph_d_release(struct dentry *dentry)
1048 struct ceph_dentry_info *di = ceph_dentry(dentry);
1050 dout("d_release %p\n", dentry);
1051 ceph_dentry_lru_del(dentry);
1052 if (di->lease_session)
1053 ceph_put_mds_session(di->lease_session);
1054 kmem_cache_free(ceph_dentry_cachep, di);
1055 dentry->d_fsdata = NULL;
1058 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1062 * Eventually, we'll want to revalidate snapped metadata
1063 * too... probably...
1069 * When the VFS prunes a dentry from the cache, we need to clear the
1070 * complete flag on the parent directory.
1072 * Called under dentry->d_lock.
1074 static void ceph_d_prune(struct dentry *dentry)
1076 dout("ceph_d_prune %p\n", dentry);
1078 /* do we have a valid parent? */
1079 if (IS_ROOT(dentry))
1082 /* if we are not hashed, we don't affect I_COMPLETE */
1083 if (d_unhashed(dentry))
1087 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1088 * cleared until d_release
1090 ceph_i_clear(dentry->d_parent->d_inode, CEPH_I_COMPLETE);
1094 * read() on a dir. This weird interface hack only works if mounted
1095 * with '-o dirstat'.
1097 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1100 struct ceph_file_info *cf = file->private_data;
1101 struct inode *inode = file_inode(file);
1102 struct ceph_inode_info *ci = ceph_inode(inode);
1104 const int bufsize = 1024;
1106 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1109 if (!cf->dir_info) {
1110 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1114 snprintf(cf->dir_info, bufsize,
1117 " subdirs: %20lld\n"
1118 "rentries: %20lld\n"
1120 " rsubdirs: %20lld\n"
1122 "rctime: %10ld.%09ld\n",
1123 ci->i_files + ci->i_subdirs,
1126 ci->i_rfiles + ci->i_rsubdirs,
1130 (long)ci->i_rctime.tv_sec,
1131 (long)ci->i_rctime.tv_nsec);
1134 if (*ppos >= cf->dir_info_len)
1136 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1137 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1140 *ppos += (size - left);
1145 * an fsync() on a dir will wait for any uncommitted directory
1146 * operations to commit.
1148 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1151 struct inode *inode = file_inode(file);
1152 struct ceph_inode_info *ci = ceph_inode(inode);
1153 struct list_head *head = &ci->i_unsafe_dirops;
1154 struct ceph_mds_request *req;
1158 dout("dir_fsync %p\n", inode);
1159 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1162 mutex_lock(&inode->i_mutex);
1164 spin_lock(&ci->i_unsafe_lock);
1165 if (list_empty(head))
1168 req = list_entry(head->prev,
1169 struct ceph_mds_request, r_unsafe_dir_item);
1170 last_tid = req->r_tid;
1173 ceph_mdsc_get_request(req);
1174 spin_unlock(&ci->i_unsafe_lock);
1176 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1177 inode, req->r_tid, last_tid);
1178 if (req->r_timeout) {
1179 ret = wait_for_completion_timeout(
1180 &req->r_safe_completion, req->r_timeout);
1184 ret = -EIO; /* timed out */
1186 wait_for_completion(&req->r_safe_completion);
1188 ceph_mdsc_put_request(req);
1190 spin_lock(&ci->i_unsafe_lock);
1191 if (ret || list_empty(head))
1193 req = list_entry(head->next,
1194 struct ceph_mds_request, r_unsafe_dir_item);
1195 } while (req->r_tid < last_tid);
1197 spin_unlock(&ci->i_unsafe_lock);
1198 mutex_unlock(&inode->i_mutex);
1204 * We maintain a private dentry LRU.
1206 * FIXME: this needs to be changed to a per-mds lru to be useful.
1208 void ceph_dentry_lru_add(struct dentry *dn)
1210 struct ceph_dentry_info *di = ceph_dentry(dn);
1211 struct ceph_mds_client *mdsc;
1213 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1214 dn->d_name.len, dn->d_name.name);
1215 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1216 spin_lock(&mdsc->dentry_lru_lock);
1217 list_add_tail(&di->lru, &mdsc->dentry_lru);
1219 spin_unlock(&mdsc->dentry_lru_lock);
1222 void ceph_dentry_lru_touch(struct dentry *dn)
1224 struct ceph_dentry_info *di = ceph_dentry(dn);
1225 struct ceph_mds_client *mdsc;
1227 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1228 dn->d_name.len, dn->d_name.name, di->offset);
1229 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1230 spin_lock(&mdsc->dentry_lru_lock);
1231 list_move_tail(&di->lru, &mdsc->dentry_lru);
1232 spin_unlock(&mdsc->dentry_lru_lock);
1235 void ceph_dentry_lru_del(struct dentry *dn)
1237 struct ceph_dentry_info *di = ceph_dentry(dn);
1238 struct ceph_mds_client *mdsc;
1240 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1241 dn->d_name.len, dn->d_name.name);
1242 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1243 spin_lock(&mdsc->dentry_lru_lock);
1244 list_del_init(&di->lru);
1246 spin_unlock(&mdsc->dentry_lru_lock);
1250 * Return name hash for a given dentry. This is dependent on
1251 * the parent directory's hash function.
1253 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1255 struct ceph_inode_info *dci = ceph_inode(dir);
1257 switch (dci->i_dir_layout.dl_dir_hash) {
1258 case 0: /* for backward compat */
1259 case CEPH_STR_HASH_LINUX:
1260 return dn->d_name.hash;
1263 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1264 dn->d_name.name, dn->d_name.len);
1268 const struct file_operations ceph_dir_fops = {
1269 .read = ceph_read_dir,
1270 .readdir = ceph_readdir,
1271 .llseek = ceph_dir_llseek,
1273 .release = ceph_release,
1274 .unlocked_ioctl = ceph_ioctl,
1275 .fsync = ceph_dir_fsync,
1278 const struct inode_operations ceph_dir_iops = {
1279 .lookup = ceph_lookup,
1280 .permission = ceph_permission,
1281 .getattr = ceph_getattr,
1282 .setattr = ceph_setattr,
1283 .setxattr = ceph_setxattr,
1284 .getxattr = ceph_getxattr,
1285 .listxattr = ceph_listxattr,
1286 .removexattr = ceph_removexattr,
1287 .mknod = ceph_mknod,
1288 .symlink = ceph_symlink,
1289 .mkdir = ceph_mkdir,
1291 .unlink = ceph_unlink,
1292 .rmdir = ceph_unlink,
1293 .rename = ceph_rename,
1294 .create = ceph_create,
1295 .atomic_open = ceph_atomic_open,
1298 const struct dentry_operations ceph_dentry_ops = {
1299 .d_revalidate = ceph_d_revalidate,
1300 .d_release = ceph_d_release,
1301 .d_prune = ceph_d_prune,
1304 const struct dentry_operations ceph_snapdir_dentry_ops = {
1305 .d_revalidate = ceph_snapdir_d_revalidate,
1306 .d_release = ceph_d_release,
1309 const struct dentry_operations ceph_snap_dentry_ops = {
1310 .d_release = ceph_d_release,
1311 .d_prune = ceph_d_prune,