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) /* lost a race */
57 di->lease_session = NULL;
58 dentry->d_fsdata = di;
59 dentry->d_time = jiffies;
60 ceph_dentry_lru_add(dentry);
62 spin_unlock(&dentry->d_lock);
69 * for readdir, we encode the directory frag and offset within that
72 static unsigned fpos_frag(loff_t p)
76 static unsigned fpos_off(loff_t p)
78 return p & 0xffffffff;
82 * When possible, we try to satisfy a readdir by peeking at the
83 * dcache. We make this work by carefully ordering dentries on
84 * d_u.d_child when we initially get results back from the MDS, and
85 * falling back to a "normal" sync readdir if any dentries in the dir
88 * I_COMPLETE tells indicates we have all dentries in the dir. It is
89 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
90 * the MDS if/when the directory is modified).
92 static int __dcache_readdir(struct file *filp,
93 void *dirent, filldir_t filldir)
95 struct inode *inode = filp->f_dentry->d_inode;
96 struct ceph_file_info *fi = filp->private_data;
97 struct dentry *parent = filp->f_dentry;
98 struct inode *dir = parent->d_inode;
100 struct dentry *dentry, *last;
101 struct ceph_dentry_info *di;
104 /* claim ref on last dentry we returned */
108 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
111 spin_lock(&dcache_lock);
113 /* start at beginning? */
114 if (filp->f_pos == 2 || (last &&
115 filp->f_pos < ceph_dentry(last)->offset)) {
116 if (list_empty(&parent->d_subdirs))
118 p = parent->d_subdirs.prev;
119 dout(" initial p %p/%p\n", p->prev, p->next);
121 p = last->d_u.d_child.prev;
125 dentry = list_entry(p, struct dentry, d_u.d_child);
126 di = ceph_dentry(dentry);
128 dout(" p %p/%p d_subdirs %p/%p\n", p->prev, p->next,
129 parent->d_subdirs.prev, parent->d_subdirs.next);
130 if (p == &parent->d_subdirs) {
134 if (!d_unhashed(dentry) && dentry->d_inode &&
135 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
136 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
137 filp->f_pos <= di->offset)
139 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
140 dentry->d_name.len, dentry->d_name.name, di->offset,
141 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
142 !dentry->d_inode ? " null" : "");
144 dentry = list_entry(p, struct dentry, d_u.d_child);
145 di = ceph_dentry(dentry);
148 atomic_inc(&dentry->d_count);
149 spin_unlock(&dcache_lock);
150 spin_unlock(&inode->i_lock);
152 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
153 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
154 filp->f_pos = di->offset;
155 err = filldir(dirent, dentry->d_name.name,
156 dentry->d_name.len, di->offset,
157 dentry->d_inode->i_ino,
158 dentry->d_inode->i_mode >> 12);
162 /* remember our position */
164 fi->next_offset = di->offset;
171 spin_lock(&inode->i_lock);
172 spin_lock(&dcache_lock);
182 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
183 if ((ceph_inode(dir)->i_ceph_flags & CEPH_I_COMPLETE))
185 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
189 spin_unlock(&dcache_lock);
192 spin_unlock(&inode->i_lock);
194 spin_lock(&inode->i_lock);
201 * make note of the last dentry we read, so we can
202 * continue at the same lexicographical point,
203 * regardless of what dir changes take place on the
206 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
209 kfree(fi->last_name);
210 fi->last_name = kmalloc(len+1, GFP_NOFS);
213 memcpy(fi->last_name, name, len);
214 fi->last_name[len] = 0;
215 dout("note_last_dentry '%s'\n", fi->last_name);
219 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
221 struct ceph_file_info *fi = filp->private_data;
222 struct inode *inode = filp->f_dentry->d_inode;
223 struct ceph_inode_info *ci = ceph_inode(inode);
224 struct ceph_client *client = ceph_inode_to_client(inode);
225 struct ceph_mds_client *mdsc = &client->mdsc;
226 unsigned frag = fpos_frag(filp->f_pos);
227 int off = fpos_off(filp->f_pos);
230 struct ceph_mds_reply_info_parsed *rinfo;
231 const int max_entries = client->mount_args->max_readdir;
233 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
237 /* always start with . and .. */
238 if (filp->f_pos == 0) {
239 /* note dir version at start of readdir so we can tell
240 * if any dentries get dropped */
241 fi->dir_release_count = ci->i_release_count;
243 dout("readdir off 0 -> '.'\n");
244 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
245 inode->i_ino, inode->i_mode >> 12) < 0)
250 if (filp->f_pos == 1) {
251 dout("readdir off 1 -> '..'\n");
252 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
253 filp->f_dentry->d_parent->d_inode->i_ino,
254 inode->i_mode >> 12) < 0)
260 /* can we use the dcache? */
261 spin_lock(&inode->i_lock);
262 if ((filp->f_pos == 2 || fi->dentry) &&
263 !ceph_test_opt(client, NOASYNCREADDIR) &&
264 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
265 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
266 err = __dcache_readdir(filp, dirent, filldir);
267 if (err != -EAGAIN) {
268 spin_unlock(&inode->i_lock);
272 spin_unlock(&inode->i_lock);
274 err = note_last_dentry(fi, fi->dentry->d_name.name,
275 fi->dentry->d_name.len);
282 /* proceed with a normal readdir */
285 /* do we have the correct frag content buffered? */
286 if (fi->frag != frag || fi->last_readdir == NULL) {
287 struct ceph_mds_request *req;
288 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
289 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
291 /* discard old result, if any */
292 if (fi->last_readdir) {
293 ceph_mdsc_put_request(fi->last_readdir);
294 fi->last_readdir = NULL;
297 /* requery frag tree, as the frag topology may have changed */
298 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
300 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
301 ceph_vinop(inode), frag, fi->last_name);
302 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
305 req->r_inode = igrab(inode);
306 req->r_dentry = dget(filp->f_dentry);
307 /* hints to request -> mds selection code */
308 req->r_direct_mode = USE_AUTH_MDS;
309 req->r_direct_hash = ceph_frag_value(frag);
310 req->r_direct_is_hash = true;
311 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
312 req->r_readdir_offset = fi->next_offset;
313 req->r_args.readdir.frag = cpu_to_le32(frag);
314 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
315 req->r_num_caps = max_entries;
316 err = ceph_mdsc_do_request(mdsc, NULL, req);
318 ceph_mdsc_put_request(req);
321 dout("readdir got and parsed readdir result=%d"
322 " on frag %x, end=%d, complete=%d\n", err, frag,
323 (int)req->r_reply_info.dir_end,
324 (int)req->r_reply_info.dir_complete);
326 if (!req->r_did_prepopulate) {
327 dout("readdir !did_prepopulate");
328 fi->dir_release_count--; /* preclude I_COMPLETE */
331 /* note next offset and last dentry name */
332 fi->offset = fi->next_offset;
333 fi->last_readdir = req;
335 if (req->r_reply_info.dir_end) {
336 kfree(fi->last_name);
337 fi->last_name = NULL;
340 rinfo = &req->r_reply_info;
341 err = note_last_dentry(fi,
342 rinfo->dir_dname[rinfo->dir_nr-1],
343 rinfo->dir_dname_len[rinfo->dir_nr-1]);
346 fi->next_offset += rinfo->dir_nr;
350 rinfo = &fi->last_readdir->r_reply_info;
351 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
352 rinfo->dir_nr, off, fi->offset);
353 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
354 u64 pos = ceph_make_fpos(frag, off);
355 struct ceph_mds_reply_inode *in =
356 rinfo->dir_in[off - fi->offset].in;
357 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
358 off, off - fi->offset, rinfo->dir_nr, pos,
359 rinfo->dir_dname_len[off - fi->offset],
360 rinfo->dir_dname[off - fi->offset], in);
362 ftype = le32_to_cpu(in->mode) >> 12;
364 rinfo->dir_dname[off - fi->offset],
365 rinfo->dir_dname_len[off - fi->offset],
367 le64_to_cpu(in->ino),
369 dout("filldir stopping us...\n");
373 filp->f_pos = pos + 1;
377 ceph_mdsc_put_request(fi->last_readdir);
378 fi->last_readdir = NULL;
383 if (!ceph_frag_is_rightmost(frag)) {
384 frag = ceph_frag_next(frag);
386 filp->f_pos = ceph_make_fpos(frag, off);
387 dout("readdir next frag is %x\n", frag);
393 * if dir_release_count still matches the dir, no dentries
394 * were released during the whole readdir, and we should have
395 * the complete dir contents in our cache.
397 spin_lock(&inode->i_lock);
398 if (ci->i_release_count == fi->dir_release_count) {
399 dout(" marking %p complete\n", inode);
400 ci->i_ceph_flags |= CEPH_I_COMPLETE;
401 ci->i_max_offset = filp->f_pos;
403 spin_unlock(&inode->i_lock);
405 dout("readdir %p filp %p done.\n", inode, filp);
409 static void reset_readdir(struct ceph_file_info *fi)
411 if (fi->last_readdir) {
412 ceph_mdsc_put_request(fi->last_readdir);
413 fi->last_readdir = NULL;
415 kfree(fi->last_name);
416 fi->next_offset = 2; /* compensate for . and .. */
424 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
426 struct ceph_file_info *fi = file->private_data;
427 struct inode *inode = file->f_mapping->host;
428 loff_t old_offset = offset;
431 mutex_lock(&inode->i_mutex);
434 offset += inode->i_size + 2; /* FIXME */
437 offset += file->f_pos;
440 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
441 if (offset != file->f_pos) {
442 file->f_pos = offset;
449 * discard buffered readdir content on seekdir(0), or
450 * seek to new frag, or seek prior to current chunk.
453 fpos_frag(offset) != fpos_frag(old_offset) ||
454 fpos_off(offset) < fi->offset) {
455 dout("dir_llseek dropping %p content\n", file);
459 /* bump dir_release_count if we did a forward seek */
460 if (offset > old_offset)
461 fi->dir_release_count--;
463 mutex_unlock(&inode->i_mutex);
468 * Process result of a lookup/open request.
470 * Mainly, make sure we return the final req->r_dentry (if it already
471 * existed) in place of the original VFS-provided dentry when they
474 * Gracefully handle the case where the MDS replies with -ENOENT and
475 * no trace (which it may do, at its discretion, e.g., if it doesn't
476 * care to issue a lease on the negative dentry).
478 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
479 struct dentry *dentry, int err)
481 struct ceph_client *client = ceph_client(dentry->d_sb);
482 struct inode *parent = dentry->d_parent->d_inode;
485 if (err == -ENOENT &&
486 ceph_vino(parent).ino != CEPH_INO_ROOT && /* no .snap in root dir */
487 strcmp(dentry->d_name.name,
488 client->mount_args->snapdir_name) == 0) {
489 struct inode *inode = ceph_get_snapdir(parent);
490 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
491 dentry, dentry->d_name.len, dentry->d_name.name, inode);
492 d_add(dentry, inode);
496 if (err == -ENOENT) {
499 if (!req->r_reply_info.head->is_dentry) {
500 dout("ENOENT and no trace, dentry %p inode %p\n",
501 dentry, dentry->d_inode);
502 if (dentry->d_inode) {
511 dentry = ERR_PTR(err);
512 else if (dentry != req->r_dentry)
513 dentry = dget(req->r_dentry); /* we got spliced */
519 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
521 return ceph_ino(inode) == CEPH_INO_ROOT &&
522 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
526 * Look up a single dir entry. If there is a lookup intent, inform
527 * the MDS so that it gets our 'caps wanted' value in a single op.
529 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
530 struct nameidata *nd)
532 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
533 struct ceph_mds_client *mdsc = &client->mdsc;
534 struct ceph_mds_request *req;
538 dout("lookup %p dentry %p '%.*s'\n",
539 dir, dentry, dentry->d_name.len, dentry->d_name.name);
541 if (dentry->d_name.len > NAME_MAX)
542 return ERR_PTR(-ENAMETOOLONG);
544 err = ceph_init_dentry(dentry);
548 /* open (but not create!) intent? */
550 (nd->flags & LOOKUP_OPEN) &&
551 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
552 !(nd->intent.open.flags & O_CREAT)) {
553 int mode = nd->intent.open.create_mode & ~current->fs->umask;
554 return ceph_lookup_open(dir, dentry, nd, mode, 1);
557 /* can we conclude ENOENT locally? */
558 if (dentry->d_inode == NULL) {
559 struct ceph_inode_info *ci = ceph_inode(dir);
560 struct ceph_dentry_info *di = ceph_dentry(dentry);
562 spin_lock(&dir->i_lock);
563 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
564 if (strncmp(dentry->d_name.name,
565 client->mount_args->snapdir_name,
566 dentry->d_name.len) &&
567 !is_root_ceph_dentry(dir, dentry) &&
568 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
569 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
570 di->offset = ci->i_max_offset++;
571 spin_unlock(&dir->i_lock);
572 dout(" dir %p complete, -ENOENT\n", dir);
574 di->lease_shared_gen = ci->i_shared_gen;
577 spin_unlock(&dir->i_lock);
580 op = ceph_snap(dir) == CEPH_SNAPDIR ?
581 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
582 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
584 return ERR_PTR(PTR_ERR(req));
585 req->r_dentry = dget(dentry);
587 /* we only need inode linkage */
588 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
589 req->r_locked_dir = dir;
590 err = ceph_mdsc_do_request(mdsc, NULL, req);
591 dentry = ceph_finish_lookup(req, dentry, err);
592 ceph_mdsc_put_request(req); /* will dput(dentry) */
593 dout("lookup result=%p\n", dentry);
598 * If we do a create but get no trace back from the MDS, follow up with
599 * a lookup (the VFS expects us to link up the provided dentry).
601 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
603 struct dentry *result = ceph_lookup(dir, dentry, NULL);
605 if (result && !IS_ERR(result)) {
607 * We created the item, then did a lookup, and found
608 * it was already linked to another inode we already
609 * had in our cache (and thus got spliced). Link our
610 * dentry to that inode, but don't hash it, just in
611 * case the VFS wants to dereference it.
613 BUG_ON(!result->d_inode);
614 d_instantiate(dentry, result->d_inode);
617 return PTR_ERR(result);
620 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
621 int mode, dev_t rdev)
623 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
624 struct ceph_mds_client *mdsc = &client->mdsc;
625 struct ceph_mds_request *req;
628 if (ceph_snap(dir) != CEPH_NOSNAP)
631 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
632 dir, dentry, mode, rdev);
633 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
638 req->r_dentry = dget(dentry);
640 req->r_locked_dir = dir;
641 req->r_args.mknod.mode = cpu_to_le32(mode);
642 req->r_args.mknod.rdev = cpu_to_le32(rdev);
643 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
644 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
645 err = ceph_mdsc_do_request(mdsc, dir, req);
646 if (!err && !req->r_reply_info.head->is_dentry)
647 err = ceph_handle_notrace_create(dir, dentry);
648 ceph_mdsc_put_request(req);
654 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
655 struct nameidata *nd)
657 dout("create in dir %p dentry %p name '%.*s'\n",
658 dir, dentry, dentry->d_name.len, dentry->d_name.name);
660 if (ceph_snap(dir) != CEPH_NOSNAP)
664 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
665 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
666 /* hrm, what should i do here if we get aliased? */
668 return PTR_ERR(dentry);
672 /* fall back to mknod */
673 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
676 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
679 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
680 struct ceph_mds_client *mdsc = &client->mdsc;
681 struct ceph_mds_request *req;
684 if (ceph_snap(dir) != CEPH_NOSNAP)
687 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
688 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
693 req->r_dentry = dget(dentry);
695 req->r_path2 = kstrdup(dest, GFP_NOFS);
696 req->r_locked_dir = dir;
697 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
698 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
699 err = ceph_mdsc_do_request(mdsc, dir, req);
700 if (!err && !req->r_reply_info.head->is_dentry)
701 err = ceph_handle_notrace_create(dir, dentry);
702 ceph_mdsc_put_request(req);
708 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
710 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
711 struct ceph_mds_client *mdsc = &client->mdsc;
712 struct ceph_mds_request *req;
716 if (ceph_snap(dir) == CEPH_SNAPDIR) {
717 /* mkdir .snap/foo is a MKSNAP */
718 op = CEPH_MDS_OP_MKSNAP;
719 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
720 dentry->d_name.len, dentry->d_name.name, dentry);
721 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
722 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
723 op = CEPH_MDS_OP_MKDIR;
727 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
733 req->r_dentry = dget(dentry);
735 req->r_locked_dir = dir;
736 req->r_args.mkdir.mode = cpu_to_le32(mode);
737 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
738 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
739 err = ceph_mdsc_do_request(mdsc, dir, req);
740 if (!err && !req->r_reply_info.head->is_dentry)
741 err = ceph_handle_notrace_create(dir, dentry);
742 ceph_mdsc_put_request(req);
749 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
750 struct dentry *dentry)
752 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
753 struct ceph_mds_client *mdsc = &client->mdsc;
754 struct ceph_mds_request *req;
757 if (ceph_snap(dir) != CEPH_NOSNAP)
760 dout("link in dir %p old_dentry %p dentry %p\n", dir,
762 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
767 req->r_dentry = dget(dentry);
769 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
770 req->r_locked_dir = dir;
771 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
772 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
773 err = ceph_mdsc_do_request(mdsc, dir, req);
776 else if (!req->r_reply_info.head->is_dentry)
777 d_instantiate(dentry, igrab(old_dentry->d_inode));
778 ceph_mdsc_put_request(req);
783 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
784 * looks like the link count will hit 0, drop any other caps (other
785 * than PIN) we don't specifically want (due to the file still being
788 static int drop_caps_for_unlink(struct inode *inode)
790 struct ceph_inode_info *ci = ceph_inode(inode);
791 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
793 spin_lock(&inode->i_lock);
794 if (inode->i_nlink == 1) {
795 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
796 ci->i_ceph_flags |= CEPH_I_NODELAY;
798 spin_unlock(&inode->i_lock);
803 * rmdir and unlink are differ only by the metadata op code
805 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
807 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
808 struct ceph_mds_client *mdsc = &client->mdsc;
809 struct inode *inode = dentry->d_inode;
810 struct ceph_mds_request *req;
814 if (ceph_snap(dir) == CEPH_SNAPDIR) {
815 /* rmdir .snap/foo is RMSNAP */
816 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
817 dentry->d_name.name, dentry);
818 op = CEPH_MDS_OP_RMSNAP;
819 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
820 dout("unlink/rmdir dir %p dn %p inode %p\n",
822 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
823 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
826 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
831 req->r_dentry = dget(dentry);
833 req->r_locked_dir = dir;
834 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
835 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
836 req->r_inode_drop = drop_caps_for_unlink(inode);
837 err = ceph_mdsc_do_request(mdsc, dir, req);
838 if (!err && !req->r_reply_info.head->is_dentry)
840 ceph_mdsc_put_request(req);
845 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
846 struct inode *new_dir, struct dentry *new_dentry)
848 struct ceph_client *client = ceph_sb_to_client(old_dir->i_sb);
849 struct ceph_mds_client *mdsc = &client->mdsc;
850 struct ceph_mds_request *req;
853 if (ceph_snap(old_dir) != ceph_snap(new_dir))
855 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
856 ceph_snap(new_dir) != CEPH_NOSNAP)
858 dout("rename dir %p dentry %p to dir %p dentry %p\n",
859 old_dir, old_dentry, new_dir, new_dentry);
860 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
863 req->r_dentry = dget(new_dentry);
865 req->r_old_dentry = dget(old_dentry);
866 req->r_locked_dir = new_dir;
867 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
868 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
869 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
870 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
871 /* release LINK_RDCACHE on source inode (mds will lock it) */
872 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
873 if (new_dentry->d_inode)
874 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
875 err = ceph_mdsc_do_request(mdsc, old_dir, req);
876 if (!err && !req->r_reply_info.head->is_dentry) {
878 * Normally d_move() is done by fill_trace (called by
879 * do_request, above). If there is no trace, we need
882 d_move(old_dentry, new_dentry);
884 ceph_mdsc_put_request(req);
890 * Check if dentry lease is valid. If not, delete the lease. Try to
891 * renew if the least is more than half up.
893 static int dentry_lease_is_valid(struct dentry *dentry)
895 struct ceph_dentry_info *di;
896 struct ceph_mds_session *s;
900 struct ceph_mds_session *session = NULL;
901 struct inode *dir = NULL;
904 spin_lock(&dentry->d_lock);
905 di = ceph_dentry(dentry);
906 if (di && di->lease_session) {
907 s = di->lease_session;
908 spin_lock(&s->s_cap_lock);
911 spin_unlock(&s->s_cap_lock);
913 if (di->lease_gen == gen &&
914 time_before(jiffies, dentry->d_time) &&
915 time_before(jiffies, ttl)) {
917 if (di->lease_renew_after &&
918 time_after(jiffies, di->lease_renew_after)) {
919 /* we should renew */
920 dir = dentry->d_parent->d_inode;
921 session = ceph_get_mds_session(s);
923 di->lease_renew_after = 0;
924 di->lease_renew_from = jiffies;
928 spin_unlock(&dentry->d_lock);
931 ceph_mdsc_lease_send_msg(session, dir, dentry,
932 CEPH_MDS_LEASE_RENEW, seq);
933 ceph_put_mds_session(session);
935 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
940 * Check if directory-wide content lease/cap is valid.
942 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
944 struct ceph_inode_info *ci = ceph_inode(dir);
945 struct ceph_dentry_info *di = ceph_dentry(dentry);
948 spin_lock(&dir->i_lock);
949 if (ci->i_shared_gen == di->lease_shared_gen)
950 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
951 spin_unlock(&dir->i_lock);
952 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
953 dir, (unsigned)ci->i_shared_gen, dentry,
954 (unsigned)di->lease_shared_gen, valid);
959 * Check if cached dentry can be trusted.
961 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
963 struct inode *dir = dentry->d_parent->d_inode;
965 dout("d_revalidate %p '%.*s' inode %p\n", dentry,
966 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
968 /* always trust cached snapped dentries, snapdir dentry */
969 if (ceph_snap(dir) != CEPH_NOSNAP) {
970 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
971 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
974 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
977 if (dentry_lease_is_valid(dentry) ||
978 dir_lease_is_valid(dir, dentry))
981 dout("d_revalidate %p invalid\n", dentry);
985 ceph_dentry_lru_touch(dentry);
990 * When a dentry is released, clear the dir I_COMPLETE if it was part
991 * of the current dir gen.
993 static void ceph_dentry_release(struct dentry *dentry)
995 struct ceph_dentry_info *di = ceph_dentry(dentry);
996 struct inode *parent_inode = dentry->d_parent->d_inode;
999 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1001 spin_lock(&parent_inode->i_lock);
1002 if (ci->i_shared_gen == di->lease_shared_gen) {
1003 dout(" clearing %p complete (d_release)\n",
1005 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1006 ci->i_release_count++;
1008 spin_unlock(&parent_inode->i_lock);
1011 ceph_dentry_lru_del(dentry);
1012 if (di->lease_session)
1013 ceph_put_mds_session(di->lease_session);
1014 kmem_cache_free(ceph_dentry_cachep, di);
1015 dentry->d_fsdata = NULL;
1019 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1020 struct nameidata *nd)
1023 * Eventually, we'll want to revalidate snapped metadata
1024 * too... probably...
1032 * read() on a dir. This weird interface hack only works if mounted
1033 * with '-o dirstat'.
1035 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1038 struct ceph_file_info *cf = file->private_data;
1039 struct inode *inode = file->f_dentry->d_inode;
1040 struct ceph_inode_info *ci = ceph_inode(inode);
1043 if (!ceph_test_opt(ceph_client(inode->i_sb), DIRSTAT))
1046 if (!cf->dir_info) {
1047 cf->dir_info = kmalloc(1024, GFP_NOFS);
1051 sprintf(cf->dir_info,
1054 " subdirs: %20lld\n"
1055 "rentries: %20lld\n"
1057 " rsubdirs: %20lld\n"
1059 "rctime: %10ld.%09ld\n",
1060 ci->i_files + ci->i_subdirs,
1063 ci->i_rfiles + ci->i_rsubdirs,
1067 (long)ci->i_rctime.tv_sec,
1068 (long)ci->i_rctime.tv_nsec);
1071 if (*ppos >= cf->dir_info_len)
1073 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1074 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1077 *ppos += (size - left);
1082 * an fsync() on a dir will wait for any uncommitted directory
1083 * operations to commit.
1085 static int ceph_dir_fsync(struct file *file, struct dentry *dentry,
1088 struct inode *inode = dentry->d_inode;
1089 struct ceph_inode_info *ci = ceph_inode(inode);
1090 struct list_head *head = &ci->i_unsafe_dirops;
1091 struct ceph_mds_request *req;
1095 dout("dir_fsync %p\n", inode);
1096 spin_lock(&ci->i_unsafe_lock);
1097 if (list_empty(head))
1100 req = list_entry(head->prev,
1101 struct ceph_mds_request, r_unsafe_dir_item);
1102 last_tid = req->r_tid;
1105 ceph_mdsc_get_request(req);
1106 spin_unlock(&ci->i_unsafe_lock);
1107 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1108 inode, req->r_tid, last_tid);
1109 if (req->r_timeout) {
1110 ret = wait_for_completion_timeout(
1111 &req->r_safe_completion, req->r_timeout);
1115 ret = -EIO; /* timed out */
1117 wait_for_completion(&req->r_safe_completion);
1119 spin_lock(&ci->i_unsafe_lock);
1120 ceph_mdsc_put_request(req);
1122 if (ret || list_empty(head))
1124 req = list_entry(head->next,
1125 struct ceph_mds_request, r_unsafe_dir_item);
1126 } while (req->r_tid < last_tid);
1128 spin_unlock(&ci->i_unsafe_lock);
1133 * We maintain a private dentry LRU.
1135 * FIXME: this needs to be changed to a per-mds lru to be useful.
1137 void ceph_dentry_lru_add(struct dentry *dn)
1139 struct ceph_dentry_info *di = ceph_dentry(dn);
1140 struct ceph_mds_client *mdsc;
1142 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1143 dn->d_name.len, dn->d_name.name);
1145 mdsc = &ceph_client(dn->d_sb)->mdsc;
1146 spin_lock(&mdsc->dentry_lru_lock);
1147 list_add_tail(&di->lru, &mdsc->dentry_lru);
1149 spin_unlock(&mdsc->dentry_lru_lock);
1153 void ceph_dentry_lru_touch(struct dentry *dn)
1155 struct ceph_dentry_info *di = ceph_dentry(dn);
1156 struct ceph_mds_client *mdsc;
1158 dout("dentry_lru_touch %p %p '%.*s'\n", di, dn,
1159 dn->d_name.len, dn->d_name.name);
1161 mdsc = &ceph_client(dn->d_sb)->mdsc;
1162 spin_lock(&mdsc->dentry_lru_lock);
1163 list_move_tail(&di->lru, &mdsc->dentry_lru);
1164 spin_unlock(&mdsc->dentry_lru_lock);
1168 void ceph_dentry_lru_del(struct dentry *dn)
1170 struct ceph_dentry_info *di = ceph_dentry(dn);
1171 struct ceph_mds_client *mdsc;
1173 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1174 dn->d_name.len, dn->d_name.name);
1176 mdsc = &ceph_client(dn->d_sb)->mdsc;
1177 spin_lock(&mdsc->dentry_lru_lock);
1178 list_del_init(&di->lru);
1180 spin_unlock(&mdsc->dentry_lru_lock);
1184 const struct file_operations ceph_dir_fops = {
1185 .read = ceph_read_dir,
1186 .readdir = ceph_readdir,
1187 .llseek = ceph_dir_llseek,
1189 .release = ceph_release,
1190 .unlocked_ioctl = ceph_ioctl,
1191 .fsync = ceph_dir_fsync,
1194 const struct inode_operations ceph_dir_iops = {
1195 .lookup = ceph_lookup,
1196 .permission = ceph_permission,
1197 .getattr = ceph_getattr,
1198 .setattr = ceph_setattr,
1199 .setxattr = ceph_setxattr,
1200 .getxattr = ceph_getxattr,
1201 .listxattr = ceph_listxattr,
1202 .removexattr = ceph_removexattr,
1203 .mknod = ceph_mknod,
1204 .symlink = ceph_symlink,
1205 .mkdir = ceph_mkdir,
1207 .unlink = ceph_unlink,
1208 .rmdir = ceph_unlink,
1209 .rename = ceph_rename,
1210 .create = ceph_create,
1213 struct dentry_operations ceph_dentry_ops = {
1214 .d_revalidate = ceph_d_revalidate,
1215 .d_release = ceph_dentry_release,
1218 struct dentry_operations ceph_snapdir_dentry_ops = {
1219 .d_revalidate = ceph_snapdir_d_revalidate,
1222 struct dentry_operations ceph_snap_dentry_ops = {