1 #include "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>
12 * Directory operations: readdir, lookup, create, link, unlink,
17 * Ceph MDS operations are specified in terms of a base ino and
18 * relative path. Thus, the client can specify an operation on a
19 * specific inode (e.g., a getattr due to fstat(2)), or as a path
20 * relative to, say, the root directory.
22 * Normally, we limit ourselves to strict inode ops (no path component)
23 * or dentry operations (a single path component relative to an ino). The
24 * exception to this is open_root_dentry(), which will open the mount
28 const struct inode_operations ceph_dir_iops;
29 const struct file_operations ceph_dir_fops;
30 struct dentry_operations ceph_dentry_ops;
33 * Initialize ceph dentry state.
35 int ceph_init_dentry(struct dentry *dentry)
37 struct ceph_dentry_info *di;
42 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
43 dentry->d_op = &ceph_dentry_ops;
44 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
45 dentry->d_op = &ceph_snapdir_dentry_ops;
47 dentry->d_op = &ceph_snap_dentry_ops;
49 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS);
51 return -ENOMEM; /* oh well */
53 spin_lock(&dentry->d_lock);
54 if (dentry->d_fsdata) {
56 kmem_cache_free(ceph_dentry_cachep, di);
60 di->lease_session = NULL;
61 dentry->d_fsdata = di;
62 dentry->d_time = jiffies;
63 ceph_dentry_lru_add(dentry);
65 spin_unlock(&dentry->d_lock);
72 * for readdir, we encode the directory frag and offset within that
75 static unsigned fpos_frag(loff_t p)
79 static unsigned fpos_off(loff_t p)
81 return p & 0xffffffff;
85 * When possible, we try to satisfy a readdir by peeking at the
86 * dcache. We make this work by carefully ordering dentries on
87 * d_u.d_child when we initially get results back from the MDS, and
88 * falling back to a "normal" sync readdir if any dentries in the dir
91 * I_COMPLETE tells indicates we have all dentries in the dir. It is
92 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
93 * the MDS if/when the directory is modified).
95 static int __dcache_readdir(struct file *filp,
96 void *dirent, filldir_t filldir)
98 struct inode *inode = filp->f_dentry->d_inode;
99 struct ceph_file_info *fi = filp->private_data;
100 struct dentry *parent = filp->f_dentry;
101 struct inode *dir = parent->d_inode;
103 struct dentry *dentry, *last;
104 struct ceph_dentry_info *di;
107 /* claim ref on last dentry we returned */
111 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
114 spin_lock(&dcache_lock);
116 /* start at beginning? */
117 if (filp->f_pos == 2 || (last &&
118 filp->f_pos < ceph_dentry(last)->offset)) {
119 if (list_empty(&parent->d_subdirs))
121 p = parent->d_subdirs.prev;
122 dout(" initial p %p/%p\n", p->prev, p->next);
124 p = last->d_u.d_child.prev;
128 dentry = list_entry(p, struct dentry, d_u.d_child);
129 di = ceph_dentry(dentry);
131 dout(" p %p/%p d_subdirs %p/%p\n", p->prev, p->next,
132 parent->d_subdirs.prev, parent->d_subdirs.next);
133 if (p == &parent->d_subdirs) {
137 if (!d_unhashed(dentry) && dentry->d_inode &&
138 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
139 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
140 filp->f_pos <= di->offset)
142 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
143 dentry->d_name.len, dentry->d_name.name, di->offset,
144 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
145 !dentry->d_inode ? " null" : "");
147 dentry = list_entry(p, struct dentry, d_u.d_child);
148 di = ceph_dentry(dentry);
151 atomic_inc(&dentry->d_count);
152 spin_unlock(&dcache_lock);
153 spin_unlock(&inode->i_lock);
155 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
156 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
157 filp->f_pos = di->offset;
158 err = filldir(dirent, dentry->d_name.name,
159 dentry->d_name.len, di->offset,
160 dentry->d_inode->i_ino,
161 dentry->d_inode->i_mode >> 12);
165 /* remember our position */
167 fi->next_offset = di->offset;
174 spin_lock(&inode->i_lock);
175 spin_lock(&dcache_lock);
185 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
186 if ((ceph_inode(dir)->i_ceph_flags & CEPH_I_COMPLETE))
188 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
192 spin_unlock(&dcache_lock);
195 spin_unlock(&inode->i_lock);
197 spin_lock(&inode->i_lock);
204 * make note of the last dentry we read, so we can
205 * continue at the same lexicographical point,
206 * regardless of what dir changes take place on the
209 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
212 kfree(fi->last_name);
213 fi->last_name = kmalloc(len+1, GFP_NOFS);
216 memcpy(fi->last_name, name, len);
217 fi->last_name[len] = 0;
218 dout("note_last_dentry '%s'\n", fi->last_name);
222 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
224 struct ceph_file_info *fi = filp->private_data;
225 struct inode *inode = filp->f_dentry->d_inode;
226 struct ceph_inode_info *ci = ceph_inode(inode);
227 struct ceph_client *client = ceph_inode_to_client(inode);
228 struct ceph_mds_client *mdsc = &client->mdsc;
229 unsigned frag = fpos_frag(filp->f_pos);
230 int off = fpos_off(filp->f_pos);
233 struct ceph_mds_reply_info_parsed *rinfo;
234 const int max_entries = client->mount_args->max_readdir;
236 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
240 /* always start with . and .. */
241 if (filp->f_pos == 0) {
242 /* note dir version at start of readdir so we can tell
243 * if any dentries get dropped */
244 fi->dir_release_count = ci->i_release_count;
246 dout("readdir off 0 -> '.'\n");
247 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
248 inode->i_ino, inode->i_mode >> 12) < 0)
253 if (filp->f_pos == 1) {
254 dout("readdir off 1 -> '..'\n");
255 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
256 filp->f_dentry->d_parent->d_inode->i_ino,
257 inode->i_mode >> 12) < 0)
263 /* can we use the dcache? */
264 spin_lock(&inode->i_lock);
265 if ((filp->f_pos == 2 || fi->dentry) &&
266 !ceph_test_opt(client, NOASYNCREADDIR) &&
267 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
268 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
269 err = __dcache_readdir(filp, dirent, filldir);
270 if (err != -EAGAIN) {
271 spin_unlock(&inode->i_lock);
275 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_num_caps = max_entries + 1;
319 err = ceph_mdsc_do_request(mdsc, NULL, req);
321 ceph_mdsc_put_request(req);
324 dout("readdir got and parsed readdir result=%d"
325 " on frag %x, end=%d, complete=%d\n", err, frag,
326 (int)req->r_reply_info.dir_end,
327 (int)req->r_reply_info.dir_complete);
329 if (!req->r_did_prepopulate) {
330 dout("readdir !did_prepopulate");
331 fi->dir_release_count--; /* preclude I_COMPLETE */
334 /* note next offset and last dentry name */
335 fi->offset = fi->next_offset;
336 fi->last_readdir = req;
338 if (req->r_reply_info.dir_end) {
339 kfree(fi->last_name);
340 fi->last_name = NULL;
343 rinfo = &req->r_reply_info;
344 err = note_last_dentry(fi,
345 rinfo->dir_dname[rinfo->dir_nr-1],
346 rinfo->dir_dname_len[rinfo->dir_nr-1]);
349 fi->next_offset += rinfo->dir_nr;
353 rinfo = &fi->last_readdir->r_reply_info;
354 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
355 rinfo->dir_nr, off, fi->offset);
356 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
357 u64 pos = ceph_make_fpos(frag, off);
358 struct ceph_mds_reply_inode *in =
359 rinfo->dir_in[off - fi->offset].in;
360 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
361 off, off - fi->offset, rinfo->dir_nr, pos,
362 rinfo->dir_dname_len[off - fi->offset],
363 rinfo->dir_dname[off - fi->offset], in);
365 ftype = le32_to_cpu(in->mode) >> 12;
367 rinfo->dir_dname[off - fi->offset],
368 rinfo->dir_dname_len[off - fi->offset],
370 le64_to_cpu(in->ino),
372 dout("filldir stopping us...\n");
376 filp->f_pos = pos + 1;
380 ceph_mdsc_put_request(fi->last_readdir);
381 fi->last_readdir = NULL;
386 if (!ceph_frag_is_rightmost(frag)) {
387 frag = ceph_frag_next(frag);
389 filp->f_pos = ceph_make_fpos(frag, off);
390 dout("readdir next frag is %x\n", frag);
396 * if dir_release_count still matches the dir, no dentries
397 * were released during the whole readdir, and we should have
398 * the complete dir contents in our cache.
400 spin_lock(&inode->i_lock);
401 if (ci->i_release_count == fi->dir_release_count) {
402 dout(" marking %p complete\n", inode);
403 ci->i_ceph_flags |= CEPH_I_COMPLETE;
404 ci->i_max_offset = filp->f_pos;
406 spin_unlock(&inode->i_lock);
408 dout("readdir %p filp %p done.\n", inode, filp);
412 static void reset_readdir(struct ceph_file_info *fi)
414 if (fi->last_readdir) {
415 ceph_mdsc_put_request(fi->last_readdir);
416 fi->last_readdir = NULL;
418 kfree(fi->last_name);
419 fi->next_offset = 2; /* compensate for . and .. */
427 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
429 struct ceph_file_info *fi = file->private_data;
430 struct inode *inode = file->f_mapping->host;
431 loff_t old_offset = offset;
434 mutex_lock(&inode->i_mutex);
437 offset += inode->i_size + 2; /* FIXME */
440 offset += file->f_pos;
443 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
444 if (offset != file->f_pos) {
445 file->f_pos = offset;
452 * discard buffered readdir content on seekdir(0), or
453 * seek to new frag, or seek prior to current chunk.
456 fpos_frag(offset) != fpos_frag(old_offset) ||
457 fpos_off(offset) < fi->offset) {
458 dout("dir_llseek dropping %p content\n", file);
462 /* bump dir_release_count if we did a forward seek */
463 if (offset > old_offset)
464 fi->dir_release_count--;
466 mutex_unlock(&inode->i_mutex);
471 * Process result of a lookup/open request.
473 * Mainly, make sure we return the final req->r_dentry (if it already
474 * existed) in place of the original VFS-provided dentry when they
477 * Gracefully handle the case where the MDS replies with -ENOENT and
478 * no trace (which it may do, at its discretion, e.g., if it doesn't
479 * care to issue a lease on the negative dentry).
481 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
482 struct dentry *dentry, int err)
484 struct ceph_client *client = ceph_sb_to_client(dentry->d_sb);
485 struct inode *parent = dentry->d_parent->d_inode;
488 if (err == -ENOENT &&
489 ceph_vino(parent).ino != CEPH_INO_ROOT && /* no .snap in root dir */
490 strcmp(dentry->d_name.name,
491 client->mount_args->snapdir_name) == 0) {
492 struct inode *inode = ceph_get_snapdir(parent);
493 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
494 dentry, dentry->d_name.len, dentry->d_name.name, inode);
495 BUG_ON(!d_unhashed(dentry));
496 d_add(dentry, inode);
500 if (err == -ENOENT) {
503 if (!req->r_reply_info.head->is_dentry) {
504 dout("ENOENT and no trace, dentry %p inode %p\n",
505 dentry, dentry->d_inode);
506 if (dentry->d_inode) {
515 dentry = ERR_PTR(err);
516 else if (dentry != req->r_dentry)
517 dentry = dget(req->r_dentry); /* we got spliced */
523 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
525 return ceph_ino(inode) == CEPH_INO_ROOT &&
526 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
530 * Look up a single dir entry. If there is a lookup intent, inform
531 * the MDS so that it gets our 'caps wanted' value in a single op.
533 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
534 struct nameidata *nd)
536 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
537 struct ceph_mds_client *mdsc = &client->mdsc;
538 struct ceph_mds_request *req;
542 dout("lookup %p dentry %p '%.*s'\n",
543 dir, dentry, dentry->d_name.len, dentry->d_name.name);
545 if (dentry->d_name.len > NAME_MAX)
546 return ERR_PTR(-ENAMETOOLONG);
548 err = ceph_init_dentry(dentry);
552 /* open (but not create!) intent? */
554 (nd->flags & LOOKUP_OPEN) &&
555 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
556 !(nd->intent.open.flags & O_CREAT)) {
557 int mode = nd->intent.open.create_mode & ~current->fs->umask;
558 return ceph_lookup_open(dir, dentry, nd, mode, 1);
561 /* can we conclude ENOENT locally? */
562 if (dentry->d_inode == NULL) {
563 struct ceph_inode_info *ci = ceph_inode(dir);
564 struct ceph_dentry_info *di = ceph_dentry(dentry);
566 spin_lock(&dir->i_lock);
567 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
568 if (strncmp(dentry->d_name.name,
569 client->mount_args->snapdir_name,
570 dentry->d_name.len) &&
571 !is_root_ceph_dentry(dir, dentry) &&
572 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
573 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
574 di->offset = ci->i_max_offset++;
575 spin_unlock(&dir->i_lock);
576 dout(" dir %p complete, -ENOENT\n", dir);
578 di->lease_shared_gen = ci->i_shared_gen;
581 spin_unlock(&dir->i_lock);
584 op = ceph_snap(dir) == CEPH_SNAPDIR ?
585 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
586 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
588 return ERR_PTR(PTR_ERR(req));
589 req->r_dentry = dget(dentry);
591 /* we only need inode linkage */
592 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
593 req->r_locked_dir = dir;
594 err = ceph_mdsc_do_request(mdsc, NULL, req);
595 dentry = ceph_finish_lookup(req, dentry, err);
596 ceph_mdsc_put_request(req); /* will dput(dentry) */
597 dout("lookup result=%p\n", dentry);
602 * If we do a create but get no trace back from the MDS, follow up with
603 * a lookup (the VFS expects us to link up the provided dentry).
605 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
607 struct dentry *result = ceph_lookup(dir, dentry, NULL);
609 if (result && !IS_ERR(result)) {
611 * We created the item, then did a lookup, and found
612 * it was already linked to another inode we already
613 * had in our cache (and thus got spliced). Link our
614 * dentry to that inode, but don't hash it, just in
615 * case the VFS wants to dereference it.
617 BUG_ON(!result->d_inode);
618 d_instantiate(dentry, result->d_inode);
621 return PTR_ERR(result);
624 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
625 int mode, dev_t rdev)
627 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
628 struct ceph_mds_client *mdsc = &client->mdsc;
629 struct ceph_mds_request *req;
632 if (ceph_snap(dir) != CEPH_NOSNAP)
635 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
636 dir, dentry, mode, rdev);
637 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
642 req->r_dentry = dget(dentry);
644 req->r_locked_dir = dir;
645 req->r_args.mknod.mode = cpu_to_le32(mode);
646 req->r_args.mknod.rdev = cpu_to_le32(rdev);
647 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
648 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
649 err = ceph_mdsc_do_request(mdsc, dir, req);
650 if (!err && !req->r_reply_info.head->is_dentry)
651 err = ceph_handle_notrace_create(dir, dentry);
652 ceph_mdsc_put_request(req);
658 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
659 struct nameidata *nd)
661 dout("create in dir %p dentry %p name '%.*s'\n",
662 dir, dentry, dentry->d_name.len, dentry->d_name.name);
664 if (ceph_snap(dir) != CEPH_NOSNAP)
668 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
669 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
670 /* hrm, what should i do here if we get aliased? */
672 return PTR_ERR(dentry);
676 /* fall back to mknod */
677 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
680 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
683 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
684 struct ceph_mds_client *mdsc = &client->mdsc;
685 struct ceph_mds_request *req;
688 if (ceph_snap(dir) != CEPH_NOSNAP)
691 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
692 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
697 req->r_dentry = dget(dentry);
699 req->r_path2 = kstrdup(dest, GFP_NOFS);
700 req->r_locked_dir = dir;
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);
712 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
714 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
715 struct ceph_mds_client *mdsc = &client->mdsc;
716 struct ceph_mds_request *req;
720 if (ceph_snap(dir) == CEPH_SNAPDIR) {
721 /* mkdir .snap/foo is a MKSNAP */
722 op = CEPH_MDS_OP_MKSNAP;
723 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
724 dentry->d_name.len, dentry->d_name.name, dentry);
725 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
726 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
727 op = CEPH_MDS_OP_MKDIR;
731 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
737 req->r_dentry = dget(dentry);
739 req->r_locked_dir = dir;
740 req->r_args.mkdir.mode = cpu_to_le32(mode);
741 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
742 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
743 err = ceph_mdsc_do_request(mdsc, dir, req);
744 if (!err && !req->r_reply_info.head->is_dentry)
745 err = ceph_handle_notrace_create(dir, dentry);
746 ceph_mdsc_put_request(req);
753 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
754 struct dentry *dentry)
756 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
757 struct ceph_mds_client *mdsc = &client->mdsc;
758 struct ceph_mds_request *req;
761 if (ceph_snap(dir) != CEPH_NOSNAP)
764 dout("link in dir %p old_dentry %p dentry %p\n", dir,
766 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
771 req->r_dentry = dget(dentry);
773 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
774 req->r_locked_dir = dir;
775 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
776 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
777 err = ceph_mdsc_do_request(mdsc, dir, req);
780 else if (!req->r_reply_info.head->is_dentry)
781 d_instantiate(dentry, igrab(old_dentry->d_inode));
782 ceph_mdsc_put_request(req);
787 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
788 * looks like the link count will hit 0, drop any other caps (other
789 * than PIN) we don't specifically want (due to the file still being
792 static int drop_caps_for_unlink(struct inode *inode)
794 struct ceph_inode_info *ci = ceph_inode(inode);
795 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
797 spin_lock(&inode->i_lock);
798 if (inode->i_nlink == 1) {
799 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
800 ci->i_ceph_flags |= CEPH_I_NODELAY;
802 spin_unlock(&inode->i_lock);
807 * rmdir and unlink are differ only by the metadata op code
809 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
811 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
812 struct ceph_mds_client *mdsc = &client->mdsc;
813 struct inode *inode = dentry->d_inode;
814 struct ceph_mds_request *req;
818 if (ceph_snap(dir) == CEPH_SNAPDIR) {
819 /* rmdir .snap/foo is RMSNAP */
820 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
821 dentry->d_name.name, dentry);
822 op = CEPH_MDS_OP_RMSNAP;
823 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
824 dout("unlink/rmdir dir %p dn %p inode %p\n",
826 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
827 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
830 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
835 req->r_dentry = dget(dentry);
837 req->r_locked_dir = dir;
838 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
839 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
840 req->r_inode_drop = drop_caps_for_unlink(inode);
841 err = ceph_mdsc_do_request(mdsc, dir, req);
842 if (!err && !req->r_reply_info.head->is_dentry)
844 ceph_mdsc_put_request(req);
849 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
850 struct inode *new_dir, struct dentry *new_dentry)
852 struct ceph_client *client = ceph_sb_to_client(old_dir->i_sb);
853 struct ceph_mds_client *mdsc = &client->mdsc;
854 struct ceph_mds_request *req;
857 if (ceph_snap(old_dir) != ceph_snap(new_dir))
859 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
860 ceph_snap(new_dir) != CEPH_NOSNAP)
862 dout("rename dir %p dentry %p to dir %p dentry %p\n",
863 old_dir, old_dentry, new_dir, new_dentry);
864 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
867 req->r_dentry = dget(new_dentry);
869 req->r_old_dentry = dget(old_dentry);
870 req->r_locked_dir = new_dir;
871 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
872 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
873 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
874 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
875 /* release LINK_RDCACHE on source inode (mds will lock it) */
876 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
877 if (new_dentry->d_inode)
878 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
879 err = ceph_mdsc_do_request(mdsc, old_dir, req);
880 if (!err && !req->r_reply_info.head->is_dentry) {
882 * Normally d_move() is done by fill_trace (called by
883 * do_request, above). If there is no trace, we need
887 /* d_move screws up d_subdirs order */
888 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
890 d_move(old_dentry, new_dentry);
892 /* ensure target dentry is invalidated, despite
893 rehashing bug in vfs_rename_dir */
894 ceph_invalidate_dentry_lease(new_dentry);
896 ceph_mdsc_put_request(req);
901 * Ensure a dentry lease will no longer revalidate.
903 void ceph_invalidate_dentry_lease(struct dentry *dentry)
905 spin_lock(&dentry->d_lock);
906 dentry->d_time = jiffies;
907 ceph_dentry(dentry)->lease_shared_gen = 0;
908 spin_unlock(&dentry->d_lock);
912 * Check if dentry lease is valid. If not, delete the lease. Try to
913 * renew if the least is more than half up.
915 static int dentry_lease_is_valid(struct dentry *dentry)
917 struct ceph_dentry_info *di;
918 struct ceph_mds_session *s;
922 struct ceph_mds_session *session = NULL;
923 struct inode *dir = NULL;
926 spin_lock(&dentry->d_lock);
927 di = ceph_dentry(dentry);
928 if (di && di->lease_session) {
929 s = di->lease_session;
930 spin_lock(&s->s_cap_lock);
933 spin_unlock(&s->s_cap_lock);
935 if (di->lease_gen == gen &&
936 time_before(jiffies, dentry->d_time) &&
937 time_before(jiffies, ttl)) {
939 if (di->lease_renew_after &&
940 time_after(jiffies, di->lease_renew_after)) {
941 /* we should renew */
942 dir = dentry->d_parent->d_inode;
943 session = ceph_get_mds_session(s);
945 di->lease_renew_after = 0;
946 di->lease_renew_from = jiffies;
950 spin_unlock(&dentry->d_lock);
953 ceph_mdsc_lease_send_msg(session, dir, dentry,
954 CEPH_MDS_LEASE_RENEW, seq);
955 ceph_put_mds_session(session);
957 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
962 * Check if directory-wide content lease/cap is valid.
964 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
966 struct ceph_inode_info *ci = ceph_inode(dir);
967 struct ceph_dentry_info *di = ceph_dentry(dentry);
970 spin_lock(&dir->i_lock);
971 if (ci->i_shared_gen == di->lease_shared_gen)
972 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
973 spin_unlock(&dir->i_lock);
974 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
975 dir, (unsigned)ci->i_shared_gen, dentry,
976 (unsigned)di->lease_shared_gen, valid);
981 * Check if cached dentry can be trusted.
983 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
985 struct inode *dir = dentry->d_parent->d_inode;
987 dout("d_revalidate %p '%.*s' inode %p\n", dentry,
988 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
990 /* always trust cached snapped dentries, snapdir dentry */
991 if (ceph_snap(dir) != CEPH_NOSNAP) {
992 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
993 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
996 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
999 if (dentry_lease_is_valid(dentry) ||
1000 dir_lease_is_valid(dir, dentry))
1003 dout("d_revalidate %p invalid\n", dentry);
1007 ceph_dentry_lru_touch(dentry);
1012 * When a dentry is released, clear the dir I_COMPLETE if it was part
1013 * of the current dir gen.
1015 static void ceph_dentry_release(struct dentry *dentry)
1017 struct ceph_dentry_info *di = ceph_dentry(dentry);
1018 struct inode *parent_inode = dentry->d_parent->d_inode;
1021 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1023 spin_lock(&parent_inode->i_lock);
1024 if (ci->i_shared_gen == di->lease_shared_gen) {
1025 dout(" clearing %p complete (d_release)\n",
1027 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1028 ci->i_release_count++;
1030 spin_unlock(&parent_inode->i_lock);
1033 ceph_dentry_lru_del(dentry);
1034 if (di->lease_session)
1035 ceph_put_mds_session(di->lease_session);
1036 kmem_cache_free(ceph_dentry_cachep, di);
1037 dentry->d_fsdata = NULL;
1041 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1042 struct nameidata *nd)
1045 * Eventually, we'll want to revalidate snapped metadata
1046 * too... probably...
1054 * read() on a dir. This weird interface hack only works if mounted
1055 * with '-o dirstat'.
1057 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1060 struct ceph_file_info *cf = file->private_data;
1061 struct inode *inode = file->f_dentry->d_inode;
1062 struct ceph_inode_info *ci = ceph_inode(inode);
1065 if (!ceph_test_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1068 if (!cf->dir_info) {
1069 cf->dir_info = kmalloc(1024, GFP_NOFS);
1073 sprintf(cf->dir_info,
1076 " subdirs: %20lld\n"
1077 "rentries: %20lld\n"
1079 " rsubdirs: %20lld\n"
1081 "rctime: %10ld.%09ld\n",
1082 ci->i_files + ci->i_subdirs,
1085 ci->i_rfiles + ci->i_rsubdirs,
1089 (long)ci->i_rctime.tv_sec,
1090 (long)ci->i_rctime.tv_nsec);
1093 if (*ppos >= cf->dir_info_len)
1095 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1096 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1099 *ppos += (size - left);
1104 * an fsync() on a dir will wait for any uncommitted directory
1105 * operations to commit.
1107 static int ceph_dir_fsync(struct file *file, struct dentry *dentry,
1110 struct inode *inode = dentry->d_inode;
1111 struct ceph_inode_info *ci = ceph_inode(inode);
1112 struct list_head *head = &ci->i_unsafe_dirops;
1113 struct ceph_mds_request *req;
1117 dout("dir_fsync %p\n", inode);
1118 spin_lock(&ci->i_unsafe_lock);
1119 if (list_empty(head))
1122 req = list_entry(head->prev,
1123 struct ceph_mds_request, r_unsafe_dir_item);
1124 last_tid = req->r_tid;
1127 ceph_mdsc_get_request(req);
1128 spin_unlock(&ci->i_unsafe_lock);
1129 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1130 inode, req->r_tid, last_tid);
1131 if (req->r_timeout) {
1132 ret = wait_for_completion_timeout(
1133 &req->r_safe_completion, req->r_timeout);
1137 ret = -EIO; /* timed out */
1139 wait_for_completion(&req->r_safe_completion);
1141 spin_lock(&ci->i_unsafe_lock);
1142 ceph_mdsc_put_request(req);
1144 if (ret || list_empty(head))
1146 req = list_entry(head->next,
1147 struct ceph_mds_request, r_unsafe_dir_item);
1148 } while (req->r_tid < last_tid);
1150 spin_unlock(&ci->i_unsafe_lock);
1155 * We maintain a private dentry LRU.
1157 * FIXME: this needs to be changed to a per-mds lru to be useful.
1159 void ceph_dentry_lru_add(struct dentry *dn)
1161 struct ceph_dentry_info *di = ceph_dentry(dn);
1162 struct ceph_mds_client *mdsc;
1164 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1165 dn->d_name.len, dn->d_name.name);
1167 mdsc = &ceph_sb_to_client(dn->d_sb)->mdsc;
1168 spin_lock(&mdsc->dentry_lru_lock);
1169 list_add_tail(&di->lru, &mdsc->dentry_lru);
1171 spin_unlock(&mdsc->dentry_lru_lock);
1175 void ceph_dentry_lru_touch(struct dentry *dn)
1177 struct ceph_dentry_info *di = ceph_dentry(dn);
1178 struct ceph_mds_client *mdsc;
1180 dout("dentry_lru_touch %p %p '%.*s'\n", di, dn,
1181 dn->d_name.len, dn->d_name.name);
1183 mdsc = &ceph_sb_to_client(dn->d_sb)->mdsc;
1184 spin_lock(&mdsc->dentry_lru_lock);
1185 list_move_tail(&di->lru, &mdsc->dentry_lru);
1186 spin_unlock(&mdsc->dentry_lru_lock);
1190 void ceph_dentry_lru_del(struct dentry *dn)
1192 struct ceph_dentry_info *di = ceph_dentry(dn);
1193 struct ceph_mds_client *mdsc;
1195 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1196 dn->d_name.len, dn->d_name.name);
1198 mdsc = &ceph_sb_to_client(dn->d_sb)->mdsc;
1199 spin_lock(&mdsc->dentry_lru_lock);
1200 list_del_init(&di->lru);
1202 spin_unlock(&mdsc->dentry_lru_lock);
1206 const struct file_operations ceph_dir_fops = {
1207 .read = ceph_read_dir,
1208 .readdir = ceph_readdir,
1209 .llseek = ceph_dir_llseek,
1211 .release = ceph_release,
1212 .unlocked_ioctl = ceph_ioctl,
1213 .fsync = ceph_dir_fsync,
1216 const struct inode_operations ceph_dir_iops = {
1217 .lookup = ceph_lookup,
1218 .permission = ceph_permission,
1219 .getattr = ceph_getattr,
1220 .setattr = ceph_setattr,
1221 .setxattr = ceph_setxattr,
1222 .getxattr = ceph_getxattr,
1223 .listxattr = ceph_listxattr,
1224 .removexattr = ceph_removexattr,
1225 .mknod = ceph_mknod,
1226 .symlink = ceph_symlink,
1227 .mkdir = ceph_mkdir,
1229 .unlink = ceph_unlink,
1230 .rmdir = ceph_unlink,
1231 .rename = ceph_rename,
1232 .create = ceph_create,
1235 struct dentry_operations ceph_dentry_ops = {
1236 .d_revalidate = ceph_d_revalidate,
1237 .d_release = ceph_dentry_release,
1240 struct dentry_operations ceph_snapdir_dentry_ops = {
1241 .d_revalidate = ceph_snapdir_d_revalidate,
1244 struct dentry_operations ceph_snap_dentry_ops = {