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 %s d_subdirs %p/%p\n", p->prev, p->next,
132 d_unhashed(dentry) ? "!hashed" : "hashed",
133 parent->d_subdirs.prev, parent->d_subdirs.next);
134 if (p == &parent->d_subdirs) {
138 if (!d_unhashed(dentry) && dentry->d_inode &&
139 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
140 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
141 filp->f_pos <= di->offset)
143 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
144 dentry->d_name.len, dentry->d_name.name, di->offset,
145 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
146 !dentry->d_inode ? " null" : "");
148 dentry = list_entry(p, struct dentry, d_u.d_child);
149 di = ceph_dentry(dentry);
152 atomic_inc(&dentry->d_count);
153 spin_unlock(&dcache_lock);
154 spin_unlock(&inode->i_lock);
156 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
157 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
158 filp->f_pos = di->offset;
159 err = filldir(dirent, dentry->d_name.name,
160 dentry->d_name.len, di->offset,
161 dentry->d_inode->i_ino,
162 dentry->d_inode->i_mode >> 12);
166 /* remember our position */
168 fi->next_offset = di->offset;
175 spin_lock(&inode->i_lock);
176 spin_lock(&dcache_lock);
186 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
187 if ((ceph_inode(dir)->i_ceph_flags & CEPH_I_COMPLETE))
189 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
193 spin_unlock(&dcache_lock);
196 spin_unlock(&inode->i_lock);
198 spin_lock(&inode->i_lock);
205 * make note of the last dentry we read, so we can
206 * continue at the same lexicographical point,
207 * regardless of what dir changes take place on the
210 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
213 kfree(fi->last_name);
214 fi->last_name = kmalloc(len+1, GFP_NOFS);
217 memcpy(fi->last_name, name, len);
218 fi->last_name[len] = 0;
219 dout("note_last_dentry '%s'\n", fi->last_name);
223 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
225 struct ceph_file_info *fi = filp->private_data;
226 struct inode *inode = filp->f_dentry->d_inode;
227 struct ceph_inode_info *ci = ceph_inode(inode);
228 struct ceph_client *client = ceph_inode_to_client(inode);
229 struct ceph_mds_client *mdsc = &client->mdsc;
230 unsigned frag = fpos_frag(filp->f_pos);
231 int off = fpos_off(filp->f_pos);
234 struct ceph_mds_reply_info_parsed *rinfo;
235 const int max_entries = client->mount_args->max_readdir;
236 const int max_bytes = client->mount_args->max_readdir_bytes;
238 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
242 /* always start with . and .. */
243 if (filp->f_pos == 0) {
244 /* note dir version at start of readdir so we can tell
245 * if any dentries get dropped */
246 fi->dir_release_count = ci->i_release_count;
248 dout("readdir off 0 -> '.'\n");
249 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
250 inode->i_ino, inode->i_mode >> 12) < 0)
255 if (filp->f_pos == 1) {
256 dout("readdir off 1 -> '..'\n");
257 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
258 filp->f_dentry->d_parent->d_inode->i_ino,
259 inode->i_mode >> 12) < 0)
265 /* can we use the dcache? */
266 spin_lock(&inode->i_lock);
267 if ((filp->f_pos == 2 || fi->dentry) &&
268 !ceph_test_opt(client, NOASYNCREADDIR) &&
269 ceph_snap(inode) != CEPH_SNAPDIR &&
270 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
271 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
272 err = __dcache_readdir(filp, dirent, filldir);
273 if (err != -EAGAIN) {
274 spin_unlock(&inode->i_lock);
278 spin_unlock(&inode->i_lock);
280 err = note_last_dentry(fi, fi->dentry->d_name.name,
281 fi->dentry->d_name.len);
288 /* proceed with a normal readdir */
291 /* do we have the correct frag content buffered? */
292 if (fi->frag != frag || fi->last_readdir == NULL) {
293 struct ceph_mds_request *req;
294 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
295 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
297 /* discard old result, if any */
298 if (fi->last_readdir) {
299 ceph_mdsc_put_request(fi->last_readdir);
300 fi->last_readdir = NULL;
303 /* requery frag tree, as the frag topology may have changed */
304 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
306 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
307 ceph_vinop(inode), frag, fi->last_name);
308 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
311 req->r_inode = igrab(inode);
312 req->r_dentry = dget(filp->f_dentry);
313 /* hints to request -> mds selection code */
314 req->r_direct_mode = USE_AUTH_MDS;
315 req->r_direct_hash = ceph_frag_value(frag);
316 req->r_direct_is_hash = true;
317 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
318 req->r_readdir_offset = fi->next_offset;
319 req->r_args.readdir.frag = cpu_to_le32(frag);
320 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
321 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
322 req->r_num_caps = max_entries + 1;
323 err = ceph_mdsc_do_request(mdsc, NULL, req);
325 ceph_mdsc_put_request(req);
328 dout("readdir got and parsed readdir result=%d"
329 " on frag %x, end=%d, complete=%d\n", err, frag,
330 (int)req->r_reply_info.dir_end,
331 (int)req->r_reply_info.dir_complete);
333 if (!req->r_did_prepopulate) {
334 dout("readdir !did_prepopulate");
335 fi->dir_release_count--; /* preclude I_COMPLETE */
338 /* note next offset and last dentry name */
339 fi->offset = fi->next_offset;
340 fi->last_readdir = req;
342 if (req->r_reply_info.dir_end) {
343 kfree(fi->last_name);
344 fi->last_name = NULL;
347 rinfo = &req->r_reply_info;
348 err = note_last_dentry(fi,
349 rinfo->dir_dname[rinfo->dir_nr-1],
350 rinfo->dir_dname_len[rinfo->dir_nr-1]);
353 fi->next_offset += rinfo->dir_nr;
357 rinfo = &fi->last_readdir->r_reply_info;
358 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
359 rinfo->dir_nr, off, fi->offset);
360 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
361 u64 pos = ceph_make_fpos(frag, off);
362 struct ceph_mds_reply_inode *in =
363 rinfo->dir_in[off - fi->offset].in;
364 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
365 off, off - fi->offset, rinfo->dir_nr, pos,
366 rinfo->dir_dname_len[off - fi->offset],
367 rinfo->dir_dname[off - fi->offset], in);
369 ftype = le32_to_cpu(in->mode) >> 12;
371 rinfo->dir_dname[off - fi->offset],
372 rinfo->dir_dname_len[off - fi->offset],
374 le64_to_cpu(in->ino),
376 dout("filldir stopping us...\n");
380 filp->f_pos = pos + 1;
384 ceph_mdsc_put_request(fi->last_readdir);
385 fi->last_readdir = NULL;
390 if (!ceph_frag_is_rightmost(frag)) {
391 frag = ceph_frag_next(frag);
393 filp->f_pos = ceph_make_fpos(frag, off);
394 dout("readdir next frag is %x\n", frag);
400 * if dir_release_count still matches the dir, no dentries
401 * were released during the whole readdir, and we should have
402 * the complete dir contents in our cache.
404 spin_lock(&inode->i_lock);
405 if (ci->i_release_count == fi->dir_release_count) {
406 dout(" marking %p complete\n", inode);
407 ci->i_ceph_flags |= CEPH_I_COMPLETE;
408 ci->i_max_offset = filp->f_pos;
410 spin_unlock(&inode->i_lock);
412 dout("readdir %p filp %p done.\n", inode, filp);
416 static void reset_readdir(struct ceph_file_info *fi)
418 if (fi->last_readdir) {
419 ceph_mdsc_put_request(fi->last_readdir);
420 fi->last_readdir = NULL;
422 kfree(fi->last_name);
423 fi->next_offset = 2; /* compensate for . and .. */
431 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
433 struct ceph_file_info *fi = file->private_data;
434 struct inode *inode = file->f_mapping->host;
435 loff_t old_offset = offset;
438 mutex_lock(&inode->i_mutex);
441 offset += inode->i_size + 2; /* FIXME */
444 offset += file->f_pos;
447 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
448 if (offset != file->f_pos) {
449 file->f_pos = offset;
456 * discard buffered readdir content on seekdir(0), or
457 * seek to new frag, or seek prior to current chunk.
460 fpos_frag(offset) != fpos_frag(old_offset) ||
461 fpos_off(offset) < fi->offset) {
462 dout("dir_llseek dropping %p content\n", file);
466 /* bump dir_release_count if we did a forward seek */
467 if (offset > old_offset)
468 fi->dir_release_count--;
470 mutex_unlock(&inode->i_mutex);
475 * Process result of a lookup/open request.
477 * Mainly, make sure we return the final req->r_dentry (if it already
478 * existed) in place of the original VFS-provided dentry when they
481 * Gracefully handle the case where the MDS replies with -ENOENT and
482 * no trace (which it may do, at its discretion, e.g., if it doesn't
483 * care to issue a lease on the negative dentry).
485 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
486 struct dentry *dentry, int err)
488 struct ceph_client *client = ceph_sb_to_client(dentry->d_sb);
489 struct inode *parent = dentry->d_parent->d_inode;
492 if (err == -ENOENT &&
493 ceph_vino(parent).ino != CEPH_INO_ROOT && /* no .snap in root dir */
494 strcmp(dentry->d_name.name,
495 client->mount_args->snapdir_name) == 0) {
496 struct inode *inode = ceph_get_snapdir(parent);
497 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
498 dentry, dentry->d_name.len, dentry->d_name.name, inode);
499 BUG_ON(!d_unhashed(dentry));
500 d_add(dentry, inode);
504 if (err == -ENOENT) {
507 if (!req->r_reply_info.head->is_dentry) {
508 dout("ENOENT and no trace, dentry %p inode %p\n",
509 dentry, dentry->d_inode);
510 if (dentry->d_inode) {
519 dentry = ERR_PTR(err);
520 else if (dentry != req->r_dentry)
521 dentry = dget(req->r_dentry); /* we got spliced */
527 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
529 return ceph_ino(inode) == CEPH_INO_ROOT &&
530 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
534 * Look up a single dir entry. If there is a lookup intent, inform
535 * the MDS so that it gets our 'caps wanted' value in a single op.
537 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
538 struct nameidata *nd)
540 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
541 struct ceph_mds_client *mdsc = &client->mdsc;
542 struct ceph_mds_request *req;
546 dout("lookup %p dentry %p '%.*s'\n",
547 dir, dentry, dentry->d_name.len, dentry->d_name.name);
549 if (dentry->d_name.len > NAME_MAX)
550 return ERR_PTR(-ENAMETOOLONG);
552 err = ceph_init_dentry(dentry);
556 /* open (but not create!) intent? */
558 (nd->flags & LOOKUP_OPEN) &&
559 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
560 !(nd->intent.open.flags & O_CREAT)) {
561 int mode = nd->intent.open.create_mode & ~current->fs->umask;
562 return ceph_lookup_open(dir, dentry, nd, mode, 1);
565 /* can we conclude ENOENT locally? */
566 if (dentry->d_inode == NULL) {
567 struct ceph_inode_info *ci = ceph_inode(dir);
568 struct ceph_dentry_info *di = ceph_dentry(dentry);
570 spin_lock(&dir->i_lock);
571 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
572 if (strncmp(dentry->d_name.name,
573 client->mount_args->snapdir_name,
574 dentry->d_name.len) &&
575 !is_root_ceph_dentry(dir, dentry) &&
576 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
577 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
578 spin_unlock(&dir->i_lock);
579 dout(" dir %p complete, -ENOENT\n", dir);
581 di->lease_shared_gen = ci->i_shared_gen;
584 spin_unlock(&dir->i_lock);
587 op = ceph_snap(dir) == CEPH_SNAPDIR ?
588 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
589 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
591 return ERR_CAST(req);
592 req->r_dentry = dget(dentry);
594 /* we only need inode linkage */
595 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
596 req->r_locked_dir = dir;
597 err = ceph_mdsc_do_request(mdsc, NULL, req);
598 dentry = ceph_finish_lookup(req, dentry, err);
599 ceph_mdsc_put_request(req); /* will dput(dentry) */
600 dout("lookup result=%p\n", dentry);
605 * If we do a create but get no trace back from the MDS, follow up with
606 * a lookup (the VFS expects us to link up the provided dentry).
608 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
610 struct dentry *result = ceph_lookup(dir, dentry, NULL);
612 if (result && !IS_ERR(result)) {
614 * We created the item, then did a lookup, and found
615 * it was already linked to another inode we already
616 * had in our cache (and thus got spliced). Link our
617 * dentry to that inode, but don't hash it, just in
618 * case the VFS wants to dereference it.
620 BUG_ON(!result->d_inode);
621 d_instantiate(dentry, result->d_inode);
624 return PTR_ERR(result);
627 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
628 int mode, dev_t rdev)
630 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
631 struct ceph_mds_client *mdsc = &client->mdsc;
632 struct ceph_mds_request *req;
635 if (ceph_snap(dir) != CEPH_NOSNAP)
638 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
639 dir, dentry, mode, rdev);
640 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
645 req->r_dentry = dget(dentry);
647 req->r_locked_dir = dir;
648 req->r_args.mknod.mode = cpu_to_le32(mode);
649 req->r_args.mknod.rdev = cpu_to_le32(rdev);
650 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
651 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
652 err = ceph_mdsc_do_request(mdsc, dir, req);
653 if (!err && !req->r_reply_info.head->is_dentry)
654 err = ceph_handle_notrace_create(dir, dentry);
655 ceph_mdsc_put_request(req);
661 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
662 struct nameidata *nd)
664 dout("create in dir %p dentry %p name '%.*s'\n",
665 dir, dentry, dentry->d_name.len, dentry->d_name.name);
667 if (ceph_snap(dir) != CEPH_NOSNAP)
671 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
672 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
673 /* hrm, what should i do here if we get aliased? */
675 return PTR_ERR(dentry);
679 /* fall back to mknod */
680 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
683 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
686 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
687 struct ceph_mds_client *mdsc = &client->mdsc;
688 struct ceph_mds_request *req;
691 if (ceph_snap(dir) != CEPH_NOSNAP)
694 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
695 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
700 req->r_dentry = dget(dentry);
702 req->r_path2 = kstrdup(dest, GFP_NOFS);
703 req->r_locked_dir = dir;
704 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
705 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
706 err = ceph_mdsc_do_request(mdsc, dir, req);
707 if (!err && !req->r_reply_info.head->is_dentry)
708 err = ceph_handle_notrace_create(dir, dentry);
709 ceph_mdsc_put_request(req);
715 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
717 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
718 struct ceph_mds_client *mdsc = &client->mdsc;
719 struct ceph_mds_request *req;
723 if (ceph_snap(dir) == CEPH_SNAPDIR) {
724 /* mkdir .snap/foo is a MKSNAP */
725 op = CEPH_MDS_OP_MKSNAP;
726 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
727 dentry->d_name.len, dentry->d_name.name, dentry);
728 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
729 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
730 op = CEPH_MDS_OP_MKDIR;
734 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
740 req->r_dentry = dget(dentry);
742 req->r_locked_dir = dir;
743 req->r_args.mkdir.mode = cpu_to_le32(mode);
744 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
745 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
746 err = ceph_mdsc_do_request(mdsc, dir, req);
747 if (!err && !req->r_reply_info.head->is_dentry)
748 err = ceph_handle_notrace_create(dir, dentry);
749 ceph_mdsc_put_request(req);
756 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
757 struct dentry *dentry)
759 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
760 struct ceph_mds_client *mdsc = &client->mdsc;
761 struct ceph_mds_request *req;
764 if (ceph_snap(dir) != CEPH_NOSNAP)
767 dout("link in dir %p old_dentry %p dentry %p\n", dir,
769 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
774 req->r_dentry = dget(dentry);
776 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
777 req->r_locked_dir = dir;
778 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
779 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
780 err = ceph_mdsc_do_request(mdsc, dir, req);
783 else if (!req->r_reply_info.head->is_dentry)
784 d_instantiate(dentry, igrab(old_dentry->d_inode));
785 ceph_mdsc_put_request(req);
790 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
791 * looks like the link count will hit 0, drop any other caps (other
792 * than PIN) we don't specifically want (due to the file still being
795 static int drop_caps_for_unlink(struct inode *inode)
797 struct ceph_inode_info *ci = ceph_inode(inode);
798 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
800 spin_lock(&inode->i_lock);
801 if (inode->i_nlink == 1) {
802 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
803 ci->i_ceph_flags |= CEPH_I_NODELAY;
805 spin_unlock(&inode->i_lock);
810 * rmdir and unlink are differ only by the metadata op code
812 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
814 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
815 struct ceph_mds_client *mdsc = &client->mdsc;
816 struct inode *inode = dentry->d_inode;
817 struct ceph_mds_request *req;
821 if (ceph_snap(dir) == CEPH_SNAPDIR) {
822 /* rmdir .snap/foo is RMSNAP */
823 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
824 dentry->d_name.name, dentry);
825 op = CEPH_MDS_OP_RMSNAP;
826 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
827 dout("unlink/rmdir dir %p dn %p inode %p\n",
829 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
830 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
833 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
838 req->r_dentry = dget(dentry);
840 req->r_locked_dir = dir;
841 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
842 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
843 req->r_inode_drop = drop_caps_for_unlink(inode);
844 err = ceph_mdsc_do_request(mdsc, dir, req);
845 if (!err && !req->r_reply_info.head->is_dentry)
847 ceph_mdsc_put_request(req);
852 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
853 struct inode *new_dir, struct dentry *new_dentry)
855 struct ceph_client *client = ceph_sb_to_client(old_dir->i_sb);
856 struct ceph_mds_client *mdsc = &client->mdsc;
857 struct ceph_mds_request *req;
860 if (ceph_snap(old_dir) != ceph_snap(new_dir))
862 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
863 ceph_snap(new_dir) != CEPH_NOSNAP)
865 dout("rename dir %p dentry %p to dir %p dentry %p\n",
866 old_dir, old_dentry, new_dir, new_dentry);
867 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
870 req->r_dentry = dget(new_dentry);
872 req->r_old_dentry = dget(old_dentry);
873 req->r_locked_dir = new_dir;
874 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
875 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
876 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
877 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
878 /* release LINK_RDCACHE on source inode (mds will lock it) */
879 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
880 if (new_dentry->d_inode)
881 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
882 err = ceph_mdsc_do_request(mdsc, old_dir, req);
883 if (!err && !req->r_reply_info.head->is_dentry) {
885 * Normally d_move() is done by fill_trace (called by
886 * do_request, above). If there is no trace, we need
890 /* d_move screws up d_subdirs order */
891 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
893 d_move(old_dentry, new_dentry);
895 /* ensure target dentry is invalidated, despite
896 rehashing bug in vfs_rename_dir */
897 ceph_invalidate_dentry_lease(new_dentry);
899 ceph_mdsc_put_request(req);
904 * Ensure a dentry lease will no longer revalidate.
906 void ceph_invalidate_dentry_lease(struct dentry *dentry)
908 spin_lock(&dentry->d_lock);
909 dentry->d_time = jiffies;
910 ceph_dentry(dentry)->lease_shared_gen = 0;
911 spin_unlock(&dentry->d_lock);
915 * Check if dentry lease is valid. If not, delete the lease. Try to
916 * renew if the least is more than half up.
918 static int dentry_lease_is_valid(struct dentry *dentry)
920 struct ceph_dentry_info *di;
921 struct ceph_mds_session *s;
925 struct ceph_mds_session *session = NULL;
926 struct inode *dir = NULL;
929 spin_lock(&dentry->d_lock);
930 di = ceph_dentry(dentry);
931 if (di && di->lease_session) {
932 s = di->lease_session;
933 spin_lock(&s->s_cap_lock);
936 spin_unlock(&s->s_cap_lock);
938 if (di->lease_gen == gen &&
939 time_before(jiffies, dentry->d_time) &&
940 time_before(jiffies, ttl)) {
942 if (di->lease_renew_after &&
943 time_after(jiffies, di->lease_renew_after)) {
944 /* we should renew */
945 dir = dentry->d_parent->d_inode;
946 session = ceph_get_mds_session(s);
948 di->lease_renew_after = 0;
949 di->lease_renew_from = jiffies;
953 spin_unlock(&dentry->d_lock);
956 ceph_mdsc_lease_send_msg(session, dir, dentry,
957 CEPH_MDS_LEASE_RENEW, seq);
958 ceph_put_mds_session(session);
960 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
965 * Check if directory-wide content lease/cap is valid.
967 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
969 struct ceph_inode_info *ci = ceph_inode(dir);
970 struct ceph_dentry_info *di = ceph_dentry(dentry);
973 spin_lock(&dir->i_lock);
974 if (ci->i_shared_gen == di->lease_shared_gen)
975 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
976 spin_unlock(&dir->i_lock);
977 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
978 dir, (unsigned)ci->i_shared_gen, dentry,
979 (unsigned)di->lease_shared_gen, valid);
984 * Check if cached dentry can be trusted.
986 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
988 struct inode *dir = dentry->d_parent->d_inode;
990 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
991 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
992 ceph_dentry(dentry)->offset);
994 /* always trust cached snapped dentries, snapdir dentry */
995 if (ceph_snap(dir) != CEPH_NOSNAP) {
996 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
997 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1000 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1003 if (dentry_lease_is_valid(dentry) ||
1004 dir_lease_is_valid(dir, dentry))
1007 dout("d_revalidate %p invalid\n", dentry);
1011 ceph_dentry_lru_touch(dentry);
1016 * When a dentry is released, clear the dir I_COMPLETE if it was part
1017 * of the current dir gen.
1019 static void ceph_dentry_release(struct dentry *dentry)
1021 struct ceph_dentry_info *di = ceph_dentry(dentry);
1022 struct inode *parent_inode = dentry->d_parent->d_inode;
1025 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1027 spin_lock(&parent_inode->i_lock);
1028 if (ci->i_shared_gen == di->lease_shared_gen) {
1029 dout(" clearing %p complete (d_release)\n",
1031 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1032 ci->i_release_count++;
1034 spin_unlock(&parent_inode->i_lock);
1037 ceph_dentry_lru_del(dentry);
1038 if (di->lease_session)
1039 ceph_put_mds_session(di->lease_session);
1040 kmem_cache_free(ceph_dentry_cachep, di);
1041 dentry->d_fsdata = NULL;
1045 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1046 struct nameidata *nd)
1049 * Eventually, we'll want to revalidate snapped metadata
1050 * too... probably...
1058 * read() on a dir. This weird interface hack only works if mounted
1059 * with '-o dirstat'.
1061 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1064 struct ceph_file_info *cf = file->private_data;
1065 struct inode *inode = file->f_dentry->d_inode;
1066 struct ceph_inode_info *ci = ceph_inode(inode);
1069 if (!ceph_test_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1072 if (!cf->dir_info) {
1073 cf->dir_info = kmalloc(1024, GFP_NOFS);
1077 sprintf(cf->dir_info,
1080 " subdirs: %20lld\n"
1081 "rentries: %20lld\n"
1083 " rsubdirs: %20lld\n"
1085 "rctime: %10ld.%09ld\n",
1086 ci->i_files + ci->i_subdirs,
1089 ci->i_rfiles + ci->i_rsubdirs,
1093 (long)ci->i_rctime.tv_sec,
1094 (long)ci->i_rctime.tv_nsec);
1097 if (*ppos >= cf->dir_info_len)
1099 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1100 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1103 *ppos += (size - left);
1108 * an fsync() on a dir will wait for any uncommitted directory
1109 * operations to commit.
1111 static int ceph_dir_fsync(struct file *file, int datasync)
1113 struct inode *inode = file->f_path.dentry->d_inode;
1114 struct ceph_inode_info *ci = ceph_inode(inode);
1115 struct list_head *head = &ci->i_unsafe_dirops;
1116 struct ceph_mds_request *req;
1120 dout("dir_fsync %p\n", inode);
1121 spin_lock(&ci->i_unsafe_lock);
1122 if (list_empty(head))
1125 req = list_entry(head->prev,
1126 struct ceph_mds_request, r_unsafe_dir_item);
1127 last_tid = req->r_tid;
1130 ceph_mdsc_get_request(req);
1131 spin_unlock(&ci->i_unsafe_lock);
1132 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1133 inode, req->r_tid, last_tid);
1134 if (req->r_timeout) {
1135 ret = wait_for_completion_timeout(
1136 &req->r_safe_completion, req->r_timeout);
1140 ret = -EIO; /* timed out */
1142 wait_for_completion(&req->r_safe_completion);
1144 spin_lock(&ci->i_unsafe_lock);
1145 ceph_mdsc_put_request(req);
1147 if (ret || list_empty(head))
1149 req = list_entry(head->next,
1150 struct ceph_mds_request, r_unsafe_dir_item);
1151 } while (req->r_tid < last_tid);
1153 spin_unlock(&ci->i_unsafe_lock);
1158 * We maintain a private dentry LRU.
1160 * FIXME: this needs to be changed to a per-mds lru to be useful.
1162 void ceph_dentry_lru_add(struct dentry *dn)
1164 struct ceph_dentry_info *di = ceph_dentry(dn);
1165 struct ceph_mds_client *mdsc;
1167 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1168 dn->d_name.len, dn->d_name.name);
1170 mdsc = &ceph_sb_to_client(dn->d_sb)->mdsc;
1171 spin_lock(&mdsc->dentry_lru_lock);
1172 list_add_tail(&di->lru, &mdsc->dentry_lru);
1174 spin_unlock(&mdsc->dentry_lru_lock);
1178 void ceph_dentry_lru_touch(struct dentry *dn)
1180 struct ceph_dentry_info *di = ceph_dentry(dn);
1181 struct ceph_mds_client *mdsc;
1183 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1184 dn->d_name.len, dn->d_name.name, di->offset);
1186 mdsc = &ceph_sb_to_client(dn->d_sb)->mdsc;
1187 spin_lock(&mdsc->dentry_lru_lock);
1188 list_move_tail(&di->lru, &mdsc->dentry_lru);
1189 spin_unlock(&mdsc->dentry_lru_lock);
1193 void ceph_dentry_lru_del(struct dentry *dn)
1195 struct ceph_dentry_info *di = ceph_dentry(dn);
1196 struct ceph_mds_client *mdsc;
1198 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1199 dn->d_name.len, dn->d_name.name);
1201 mdsc = &ceph_sb_to_client(dn->d_sb)->mdsc;
1202 spin_lock(&mdsc->dentry_lru_lock);
1203 list_del_init(&di->lru);
1205 spin_unlock(&mdsc->dentry_lru_lock);
1209 const struct file_operations ceph_dir_fops = {
1210 .read = ceph_read_dir,
1211 .readdir = ceph_readdir,
1212 .llseek = ceph_dir_llseek,
1214 .release = ceph_release,
1215 .unlocked_ioctl = ceph_ioctl,
1216 .fsync = ceph_dir_fsync,
1219 const struct inode_operations ceph_dir_iops = {
1220 .lookup = ceph_lookup,
1221 .permission = ceph_permission,
1222 .getattr = ceph_getattr,
1223 .setattr = ceph_setattr,
1224 .setxattr = ceph_setxattr,
1225 .getxattr = ceph_getxattr,
1226 .listxattr = ceph_listxattr,
1227 .removexattr = ceph_removexattr,
1228 .mknod = ceph_mknod,
1229 .symlink = ceph_symlink,
1230 .mkdir = ceph_mkdir,
1232 .unlink = ceph_unlink,
1233 .rmdir = ceph_unlink,
1234 .rename = ceph_rename,
1235 .create = ceph_create,
1238 struct dentry_operations ceph_dentry_ops = {
1239 .d_revalidate = ceph_d_revalidate,
1240 .d_release = ceph_dentry_release,
1243 struct dentry_operations ceph_snapdir_dentry_ops = {
1244 .d_revalidate = ceph_snapdir_d_revalidate,
1247 struct dentry_operations ceph_snap_dentry_ops = {