1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/spinlock.h>
4 #include <linux/fs_struct.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
18 * Ceph MDS operations are specified in terms of a base ino and
19 * relative path. Thus, the client can specify an operation on a
20 * specific inode (e.g., a getattr due to fstat(2)), or as a path
21 * relative to, say, the root directory.
23 * Normally, we limit ourselves to strict inode ops (no path component)
24 * or dentry operations (a single path component relative to an ino). The
25 * exception to this is open_root_dentry(), which will open the mount
29 const struct dentry_operations ceph_dentry_ops;
32 * Initialize ceph dentry state.
34 int ceph_init_dentry(struct dentry *dentry)
36 struct ceph_dentry_info *di;
41 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
43 return -ENOMEM; /* oh well */
45 spin_lock(&dentry->d_lock);
46 if (dentry->d_fsdata) {
48 kmem_cache_free(ceph_dentry_cachep, di);
52 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP)
53 d_set_d_op(dentry, &ceph_dentry_ops);
54 else if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
55 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
57 d_set_d_op(dentry, &ceph_snap_dentry_ops);
60 di->lease_session = NULL;
61 dentry->d_time = jiffies;
62 /* avoid reordering d_fsdata setup so that the check above is safe */
64 dentry->d_fsdata = di;
65 ceph_dentry_lru_add(dentry);
67 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;
84 static int fpos_cmp(loff_t l, loff_t r)
86 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
89 return (int)(fpos_off(l) - fpos_off(r));
93 * make note of the last dentry we read, so we can
94 * continue at the same lexicographical point,
95 * regardless of what dir changes take place on the
98 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
99 int len, unsigned next_offset)
101 char *buf = kmalloc(len+1, GFP_KERNEL);
104 kfree(fi->last_name);
106 memcpy(fi->last_name, name, len);
107 fi->last_name[len] = 0;
108 fi->next_offset = next_offset;
109 dout("note_last_dentry '%s'\n", fi->last_name);
114 * When possible, we try to satisfy a readdir by peeking at the
115 * dcache. We make this work by carefully ordering dentries on
116 * d_child when we initially get results back from the MDS, and
117 * falling back to a "normal" sync readdir if any dentries in the dir
120 * Complete dir indicates that we have all dentries in the dir. It is
121 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
122 * the MDS if/when the directory is modified).
124 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
127 struct ceph_file_info *fi = file->private_data;
128 struct dentry *parent = file->f_path.dentry;
129 struct inode *dir = d_inode(parent);
130 struct dentry *dentry, *last = NULL;
131 struct ceph_dentry_info *di;
132 unsigned nsize = PAGE_SIZE / sizeof(struct dentry *);
135 struct ceph_readdir_cache_control cache_ctl = {};
137 dout("__dcache_readdir %p v%u at %llu\n", dir, shared_gen, ctx->pos);
139 /* we can calculate cache index for the first dirfrag */
140 if (ceph_frag_is_leftmost(fpos_frag(ctx->pos))) {
141 cache_ctl.index = fpos_off(ctx->pos) - 2;
142 BUG_ON(cache_ctl.index < 0);
143 ptr_pos = cache_ctl.index * sizeof(struct dentry *);
150 if (ptr_pos >= i_size_read(dir)) {
151 fi->flags |= CEPH_F_ATEND;
157 pgoff = ptr_pos >> PAGE_SHIFT;
158 if (!cache_ctl.page || pgoff != page_index(cache_ctl.page)) {
159 ceph_readdir_cache_release(&cache_ctl);
160 cache_ctl.page = find_lock_page(&dir->i_data, pgoff);
161 if (!cache_ctl.page) {
162 dout(" page %lu not found\n", pgoff);
165 /* reading/filling the cache are serialized by
166 * i_mutex, no need to use page lock */
167 unlock_page(cache_ctl.page);
168 cache_ctl.dentries = kmap(cache_ctl.page);
172 spin_lock(&parent->d_lock);
173 /* check i_size again here, because empty directory can be
174 * marked as complete while not holding the i_mutex. */
175 if (ceph_dir_is_complete_ordered(dir) &&
176 ptr_pos < i_size_read(dir))
177 dentry = cache_ctl.dentries[cache_ctl.index % nsize];
180 spin_unlock(&parent->d_lock);
181 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
188 di = ceph_dentry(dentry);
189 spin_lock(&dentry->d_lock);
190 if (di->lease_shared_gen == shared_gen &&
191 d_really_is_positive(dentry) &&
192 ceph_snap(d_inode(dentry)) != CEPH_SNAPDIR &&
193 ceph_ino(d_inode(dentry)) != CEPH_INO_CEPH &&
194 fpos_cmp(ctx->pos, di->offset) <= 0) {
197 spin_unlock(&dentry->d_lock);
200 dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
201 dentry, dentry, d_inode(dentry));
202 ctx->pos = di->offset;
203 if (!dir_emit(ctx, dentry->d_name.name,
205 ceph_translate_ino(dentry->d_sb,
206 d_inode(dentry)->i_ino),
207 d_inode(dentry)->i_mode >> 12)) {
222 ptr_pos += sizeof(struct dentry *);
224 ceph_readdir_cache_release(&cache_ctl);
227 di = ceph_dentry(last);
228 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
229 fpos_off(di->offset) + 1);
237 static int ceph_readdir(struct file *file, struct dir_context *ctx)
239 struct ceph_file_info *fi = file->private_data;
240 struct inode *inode = file_inode(file);
241 struct ceph_inode_info *ci = ceph_inode(inode);
242 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
243 struct ceph_mds_client *mdsc = fsc->mdsc;
244 unsigned frag = fpos_frag(ctx->pos);
245 int off = fpos_off(ctx->pos);
248 struct ceph_mds_reply_info_parsed *rinfo;
250 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
251 if (fi->flags & CEPH_F_ATEND)
254 /* always start with . and .. */
256 dout("readdir off 0 -> '.'\n");
257 if (!dir_emit(ctx, ".", 1,
258 ceph_translate_ino(inode->i_sb, inode->i_ino),
259 inode->i_mode >> 12))
265 ino_t ino = parent_ino(file->f_path.dentry);
266 dout("readdir off 1 -> '..'\n");
267 if (!dir_emit(ctx, "..", 2,
268 ceph_translate_ino(inode->i_sb, ino),
269 inode->i_mode >> 12))
275 /* can we use the dcache? */
276 spin_lock(&ci->i_ceph_lock);
277 if (ceph_test_mount_opt(fsc, DCACHE) &&
278 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
279 ceph_snap(inode) != CEPH_SNAPDIR &&
280 __ceph_dir_is_complete_ordered(ci) &&
281 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
282 u32 shared_gen = ci->i_shared_gen;
283 spin_unlock(&ci->i_ceph_lock);
284 err = __dcache_readdir(file, ctx, shared_gen);
287 frag = fpos_frag(ctx->pos);
288 off = fpos_off(ctx->pos);
290 spin_unlock(&ci->i_ceph_lock);
293 /* proceed with a normal readdir */
295 /* do we have the correct frag content buffered? */
296 if (fi->frag != frag || fi->last_readdir == NULL) {
297 struct ceph_mds_request *req;
298 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
299 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
301 /* discard old result, if any */
302 if (fi->last_readdir) {
303 ceph_mdsc_put_request(fi->last_readdir);
304 fi->last_readdir = NULL;
307 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
308 ceph_vinop(inode), frag, fi->last_name);
309 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
312 err = ceph_alloc_readdir_reply_buffer(req, inode);
314 ceph_mdsc_put_request(req);
317 /* hints to request -> mds selection code */
318 req->r_direct_mode = USE_AUTH_MDS;
319 req->r_direct_hash = ceph_frag_value(frag);
320 req->r_direct_is_hash = true;
322 req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
324 ceph_mdsc_put_request(req);
328 req->r_dir_release_cnt = fi->dir_release_count;
329 req->r_dir_ordered_cnt = fi->dir_ordered_count;
330 req->r_readdir_cache_idx = fi->readdir_cache_idx;
331 req->r_readdir_offset = fi->next_offset;
332 req->r_args.readdir.frag = cpu_to_le32(frag);
334 req->r_inode = inode;
336 req->r_dentry = dget(file->f_path.dentry);
337 err = ceph_mdsc_do_request(mdsc, NULL, req);
339 ceph_mdsc_put_request(req);
342 dout("readdir got and parsed readdir result=%d"
343 " on frag %x, end=%d, complete=%d\n", err, frag,
344 (int)req->r_reply_info.dir_end,
345 (int)req->r_reply_info.dir_complete);
348 /* note next offset and last dentry name */
349 rinfo = &req->r_reply_info;
350 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
351 frag = le32_to_cpu(rinfo->dir_dir->frag);
352 off = req->r_readdir_offset;
353 fi->next_offset = off;
357 fi->offset = fi->next_offset;
358 fi->last_readdir = req;
360 if (req->r_did_prepopulate) {
361 fi->readdir_cache_idx = req->r_readdir_cache_idx;
362 if (fi->readdir_cache_idx < 0) {
363 /* preclude from marking dir ordered */
364 fi->dir_ordered_count = 0;
365 } else if (ceph_frag_is_leftmost(frag) && off == 2) {
366 /* note dir version at start of readdir so
367 * we can tell if any dentries get dropped */
368 fi->dir_release_count = req->r_dir_release_cnt;
369 fi->dir_ordered_count = req->r_dir_ordered_cnt;
372 dout("readdir !did_prepopulate");
373 /* disable readdir cache */
374 fi->readdir_cache_idx = -1;
375 /* preclude from marking dir complete */
376 fi->dir_release_count = 0;
379 if (req->r_reply_info.dir_end) {
380 kfree(fi->last_name);
381 fi->last_name = NULL;
382 if (ceph_frag_is_rightmost(frag))
387 err = note_last_dentry(fi,
388 rinfo->dir_dname[rinfo->dir_nr-1],
389 rinfo->dir_dname_len[rinfo->dir_nr-1],
390 fi->next_offset + rinfo->dir_nr);
396 rinfo = &fi->last_readdir->r_reply_info;
397 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
398 rinfo->dir_nr, off, fi->offset);
400 ctx->pos = ceph_make_fpos(frag, off);
401 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
402 struct ceph_mds_reply_inode *in =
403 rinfo->dir_in[off - fi->offset].in;
404 struct ceph_vino vino;
407 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
408 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
409 rinfo->dir_dname_len[off - fi->offset],
410 rinfo->dir_dname[off - fi->offset], in);
412 ftype = le32_to_cpu(in->mode) >> 12;
413 vino.ino = le64_to_cpu(in->ino);
414 vino.snap = le64_to_cpu(in->snapid);
415 ino = ceph_vino_to_ino(vino);
417 rinfo->dir_dname[off - fi->offset],
418 rinfo->dir_dname_len[off - fi->offset],
419 ceph_translate_ino(inode->i_sb, ino), ftype)) {
420 dout("filldir stopping us...\n");
428 ceph_mdsc_put_request(fi->last_readdir);
429 fi->last_readdir = NULL;
434 if (!ceph_frag_is_rightmost(frag)) {
435 frag = ceph_frag_next(frag);
437 ctx->pos = ceph_make_fpos(frag, off);
438 dout("readdir next frag is %x\n", frag);
441 fi->flags |= CEPH_F_ATEND;
444 * if dir_release_count still matches the dir, no dentries
445 * were released during the whole readdir, and we should have
446 * the complete dir contents in our cache.
448 if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
449 spin_lock(&ci->i_ceph_lock);
450 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
451 dout(" marking %p complete and ordered\n", inode);
452 /* use i_size to track number of entries in
454 BUG_ON(fi->readdir_cache_idx < 0);
455 i_size_write(inode, fi->readdir_cache_idx *
456 sizeof(struct dentry*));
458 dout(" marking %p complete\n", inode);
460 __ceph_dir_set_complete(ci, fi->dir_release_count,
461 fi->dir_ordered_count);
462 spin_unlock(&ci->i_ceph_lock);
465 dout("readdir %p file %p done.\n", inode, file);
469 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
471 if (fi->last_readdir) {
472 ceph_mdsc_put_request(fi->last_readdir);
473 fi->last_readdir = NULL;
475 kfree(fi->last_name);
476 fi->last_name = NULL;
477 fi->dir_release_count = 0;
478 fi->readdir_cache_idx = -1;
479 if (ceph_frag_is_leftmost(frag))
480 fi->next_offset = 2; /* compensate for . and .. */
483 fi->flags &= ~CEPH_F_ATEND;
486 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
488 struct ceph_file_info *fi = file->private_data;
489 struct inode *inode = file->f_mapping->host;
490 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
497 offset += file->f_pos;
501 retval = -EOPNOTSUPP;
507 if (offset != file->f_pos) {
508 file->f_pos = offset;
510 fi->flags &= ~CEPH_F_ATEND;
515 fpos_frag(offset) != fi->frag ||
516 fpos_off(offset) < fi->offset) {
517 /* discard buffered readdir content on seekdir(0), or
518 * seek to new frag, or seek prior to current chunk */
519 dout("dir_llseek dropping %p content\n", file);
520 reset_readdir(fi, fpos_frag(offset));
521 } else if (fpos_cmp(offset, old_offset) > 0) {
522 /* reset dir_release_count if we did a forward seek */
523 fi->dir_release_count = 0;
524 fi->readdir_cache_idx = -1;
533 * Handle lookups for the hidden .snap directory.
535 int ceph_handle_snapdir(struct ceph_mds_request *req,
536 struct dentry *dentry, int err)
538 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
539 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
542 if (err == -ENOENT &&
543 ceph_snap(parent) == CEPH_NOSNAP &&
544 strcmp(dentry->d_name.name,
545 fsc->mount_options->snapdir_name) == 0) {
546 struct inode *inode = ceph_get_snapdir(parent);
547 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
548 dentry, dentry, inode);
549 BUG_ON(!d_unhashed(dentry));
550 d_add(dentry, inode);
557 * Figure out final result of a lookup/open request.
559 * Mainly, make sure we return the final req->r_dentry (if it already
560 * existed) in place of the original VFS-provided dentry when they
563 * Gracefully handle the case where the MDS replies with -ENOENT and
564 * no trace (which it may do, at its discretion, e.g., if it doesn't
565 * care to issue a lease on the negative dentry).
567 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
568 struct dentry *dentry, int err)
570 if (err == -ENOENT) {
573 if (!req->r_reply_info.head->is_dentry) {
574 dout("ENOENT and no trace, dentry %p inode %p\n",
575 dentry, d_inode(dentry));
576 if (d_really_is_positive(dentry)) {
585 dentry = ERR_PTR(err);
586 else if (dentry != req->r_dentry)
587 dentry = dget(req->r_dentry); /* we got spliced */
593 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
595 return ceph_ino(inode) == CEPH_INO_ROOT &&
596 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
600 * Look up a single dir entry. If there is a lookup intent, inform
601 * the MDS so that it gets our 'caps wanted' value in a single op.
603 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
606 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
607 struct ceph_mds_client *mdsc = fsc->mdsc;
608 struct ceph_mds_request *req;
613 dout("lookup %p dentry %p '%pd'\n",
614 dir, dentry, dentry);
616 if (dentry->d_name.len > NAME_MAX)
617 return ERR_PTR(-ENAMETOOLONG);
619 err = ceph_init_dentry(dentry);
623 /* can we conclude ENOENT locally? */
624 if (d_really_is_negative(dentry)) {
625 struct ceph_inode_info *ci = ceph_inode(dir);
626 struct ceph_dentry_info *di = ceph_dentry(dentry);
628 spin_lock(&ci->i_ceph_lock);
629 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
630 if (strncmp(dentry->d_name.name,
631 fsc->mount_options->snapdir_name,
632 dentry->d_name.len) &&
633 !is_root_ceph_dentry(dir, dentry) &&
634 ceph_test_mount_opt(fsc, DCACHE) &&
635 __ceph_dir_is_complete(ci) &&
636 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
637 spin_unlock(&ci->i_ceph_lock);
638 dout(" dir %p complete, -ENOENT\n", dir);
640 di->lease_shared_gen = ci->i_shared_gen;
643 spin_unlock(&ci->i_ceph_lock);
646 op = ceph_snap(dir) == CEPH_SNAPDIR ?
647 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
648 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
650 return ERR_CAST(req);
651 req->r_dentry = dget(dentry);
654 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
655 if (ceph_security_xattr_wanted(dir))
656 mask |= CEPH_CAP_XATTR_SHARED;
657 req->r_args.getattr.mask = cpu_to_le32(mask);
659 req->r_locked_dir = dir;
660 err = ceph_mdsc_do_request(mdsc, NULL, req);
661 err = ceph_handle_snapdir(req, dentry, err);
662 dentry = ceph_finish_lookup(req, dentry, err);
663 ceph_mdsc_put_request(req); /* will dput(dentry) */
664 dout("lookup result=%p\n", dentry);
669 * If we do a create but get no trace back from the MDS, follow up with
670 * a lookup (the VFS expects us to link up the provided dentry).
672 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
674 struct dentry *result = ceph_lookup(dir, dentry, 0);
676 if (result && !IS_ERR(result)) {
678 * We created the item, then did a lookup, and found
679 * it was already linked to another inode we already
680 * had in our cache (and thus got spliced). To not
681 * confuse VFS (especially when inode is a directory),
682 * we don't link our dentry to that inode, return an
685 * This event should be rare and it happens only when
686 * we talk to old MDS. Recent MDS does not send traceless
687 * reply for request that creates new inode.
692 return PTR_ERR(result);
695 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
696 umode_t mode, dev_t rdev)
698 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
699 struct ceph_mds_client *mdsc = fsc->mdsc;
700 struct ceph_mds_request *req;
701 struct ceph_acls_info acls = {};
704 if (ceph_snap(dir) != CEPH_NOSNAP)
707 err = ceph_pre_init_acls(dir, &mode, &acls);
711 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
712 dir, dentry, mode, rdev);
713 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
718 req->r_dentry = dget(dentry);
720 req->r_locked_dir = dir;
721 req->r_args.mknod.mode = cpu_to_le32(mode);
722 req->r_args.mknod.rdev = cpu_to_le32(rdev);
723 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
724 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
726 req->r_pagelist = acls.pagelist;
727 acls.pagelist = NULL;
729 err = ceph_mdsc_do_request(mdsc, dir, req);
730 if (!err && !req->r_reply_info.head->is_dentry)
731 err = ceph_handle_notrace_create(dir, dentry);
732 ceph_mdsc_put_request(req);
735 ceph_init_inode_acls(d_inode(dentry), &acls);
738 ceph_release_acls_info(&acls);
742 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
745 return ceph_mknod(dir, dentry, mode, 0);
748 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
751 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
752 struct ceph_mds_client *mdsc = fsc->mdsc;
753 struct ceph_mds_request *req;
756 if (ceph_snap(dir) != CEPH_NOSNAP)
759 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
760 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
765 req->r_path2 = kstrdup(dest, GFP_KERNEL);
768 ceph_mdsc_put_request(req);
771 req->r_locked_dir = dir;
772 req->r_dentry = dget(dentry);
774 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
775 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
776 err = ceph_mdsc_do_request(mdsc, dir, req);
777 if (!err && !req->r_reply_info.head->is_dentry)
778 err = ceph_handle_notrace_create(dir, dentry);
779 ceph_mdsc_put_request(req);
786 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
788 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
789 struct ceph_mds_client *mdsc = fsc->mdsc;
790 struct ceph_mds_request *req;
791 struct ceph_acls_info acls = {};
795 if (ceph_snap(dir) == CEPH_SNAPDIR) {
796 /* mkdir .snap/foo is a MKSNAP */
797 op = CEPH_MDS_OP_MKSNAP;
798 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
800 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
801 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
802 op = CEPH_MDS_OP_MKDIR;
808 err = ceph_pre_init_acls(dir, &mode, &acls);
812 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
818 req->r_dentry = dget(dentry);
820 req->r_locked_dir = dir;
821 req->r_args.mkdir.mode = cpu_to_le32(mode);
822 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
823 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
825 req->r_pagelist = acls.pagelist;
826 acls.pagelist = NULL;
828 err = ceph_mdsc_do_request(mdsc, dir, req);
830 !req->r_reply_info.head->is_target &&
831 !req->r_reply_info.head->is_dentry)
832 err = ceph_handle_notrace_create(dir, dentry);
833 ceph_mdsc_put_request(req);
836 ceph_init_inode_acls(d_inode(dentry), &acls);
839 ceph_release_acls_info(&acls);
843 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
844 struct dentry *dentry)
846 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
847 struct ceph_mds_client *mdsc = fsc->mdsc;
848 struct ceph_mds_request *req;
851 if (ceph_snap(dir) != CEPH_NOSNAP)
854 dout("link in dir %p old_dentry %p dentry %p\n", dir,
856 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
861 req->r_dentry = dget(dentry);
863 req->r_old_dentry = dget(old_dentry);
864 req->r_locked_dir = dir;
865 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
866 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
867 /* release LINK_SHARED on source inode (mds will lock it) */
868 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
869 err = ceph_mdsc_do_request(mdsc, dir, req);
872 } else if (!req->r_reply_info.head->is_dentry) {
873 ihold(d_inode(old_dentry));
874 d_instantiate(dentry, d_inode(old_dentry));
876 ceph_mdsc_put_request(req);
881 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
882 * looks like the link count will hit 0, drop any other caps (other
883 * than PIN) we don't specifically want (due to the file still being
886 static int drop_caps_for_unlink(struct inode *inode)
888 struct ceph_inode_info *ci = ceph_inode(inode);
889 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
891 spin_lock(&ci->i_ceph_lock);
892 if (inode->i_nlink == 1) {
893 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
894 ci->i_ceph_flags |= CEPH_I_NODELAY;
896 spin_unlock(&ci->i_ceph_lock);
901 * rmdir and unlink are differ only by the metadata op code
903 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
905 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
906 struct ceph_mds_client *mdsc = fsc->mdsc;
907 struct inode *inode = d_inode(dentry);
908 struct ceph_mds_request *req;
912 if (ceph_snap(dir) == CEPH_SNAPDIR) {
913 /* rmdir .snap/foo is RMSNAP */
914 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
915 op = CEPH_MDS_OP_RMSNAP;
916 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
917 dout("unlink/rmdir dir %p dn %p inode %p\n",
919 op = d_is_dir(dentry) ?
920 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
923 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
928 req->r_dentry = dget(dentry);
930 req->r_locked_dir = dir;
931 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
932 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
933 req->r_inode_drop = drop_caps_for_unlink(inode);
934 err = ceph_mdsc_do_request(mdsc, dir, req);
935 if (!err && !req->r_reply_info.head->is_dentry)
937 ceph_mdsc_put_request(req);
942 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
943 struct inode *new_dir, struct dentry *new_dentry)
945 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
946 struct ceph_mds_client *mdsc = fsc->mdsc;
947 struct ceph_mds_request *req;
948 int op = CEPH_MDS_OP_RENAME;
951 if (ceph_snap(old_dir) != ceph_snap(new_dir))
953 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
954 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
955 op = CEPH_MDS_OP_RENAMESNAP;
959 dout("rename dir %p dentry %p to dir %p dentry %p\n",
960 old_dir, old_dentry, new_dir, new_dentry);
961 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
965 req->r_dentry = dget(new_dentry);
967 req->r_old_dentry = dget(old_dentry);
968 req->r_old_dentry_dir = old_dir;
969 req->r_locked_dir = new_dir;
970 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
971 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
972 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
973 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
974 /* release LINK_RDCACHE on source inode (mds will lock it) */
975 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
976 if (d_really_is_positive(new_dentry))
977 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
978 err = ceph_mdsc_do_request(mdsc, old_dir, req);
979 if (!err && !req->r_reply_info.head->is_dentry) {
981 * Normally d_move() is done by fill_trace (called by
982 * do_request, above). If there is no trace, we need
986 /* d_move screws up sibling dentries' offsets */
987 ceph_dir_clear_complete(old_dir);
988 ceph_dir_clear_complete(new_dir);
990 d_move(old_dentry, new_dentry);
992 /* ensure target dentry is invalidated, despite
993 rehashing bug in vfs_rename_dir */
994 ceph_invalidate_dentry_lease(new_dentry);
996 ceph_mdsc_put_request(req);
1001 * Ensure a dentry lease will no longer revalidate.
1003 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1005 spin_lock(&dentry->d_lock);
1006 dentry->d_time = jiffies;
1007 ceph_dentry(dentry)->lease_shared_gen = 0;
1008 spin_unlock(&dentry->d_lock);
1012 * Check if dentry lease is valid. If not, delete the lease. Try to
1013 * renew if the least is more than half up.
1015 static int dentry_lease_is_valid(struct dentry *dentry)
1017 struct ceph_dentry_info *di;
1018 struct ceph_mds_session *s;
1022 struct ceph_mds_session *session = NULL;
1023 struct inode *dir = NULL;
1026 spin_lock(&dentry->d_lock);
1027 di = ceph_dentry(dentry);
1028 if (di->lease_session) {
1029 s = di->lease_session;
1030 spin_lock(&s->s_gen_ttl_lock);
1033 spin_unlock(&s->s_gen_ttl_lock);
1035 if (di->lease_gen == gen &&
1036 time_before(jiffies, dentry->d_time) &&
1037 time_before(jiffies, ttl)) {
1039 if (di->lease_renew_after &&
1040 time_after(jiffies, di->lease_renew_after)) {
1041 /* we should renew */
1042 dir = d_inode(dentry->d_parent);
1043 session = ceph_get_mds_session(s);
1044 seq = di->lease_seq;
1045 di->lease_renew_after = 0;
1046 di->lease_renew_from = jiffies;
1050 spin_unlock(&dentry->d_lock);
1053 ceph_mdsc_lease_send_msg(session, dir, dentry,
1054 CEPH_MDS_LEASE_RENEW, seq);
1055 ceph_put_mds_session(session);
1057 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1062 * Check if directory-wide content lease/cap is valid.
1064 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1066 struct ceph_inode_info *ci = ceph_inode(dir);
1067 struct ceph_dentry_info *di = ceph_dentry(dentry);
1070 spin_lock(&ci->i_ceph_lock);
1071 if (ci->i_shared_gen == di->lease_shared_gen)
1072 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1073 spin_unlock(&ci->i_ceph_lock);
1074 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1075 dir, (unsigned)ci->i_shared_gen, dentry,
1076 (unsigned)di->lease_shared_gen, valid);
1081 * Check if cached dentry can be trusted.
1083 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1086 struct dentry *parent;
1089 if (flags & LOOKUP_RCU)
1092 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1093 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1095 parent = dget_parent(dentry);
1096 dir = d_inode(parent);
1098 /* always trust cached snapped dentries, snapdir dentry */
1099 if (ceph_snap(dir) != CEPH_NOSNAP) {
1100 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1101 dentry, d_inode(dentry));
1103 } else if (d_really_is_positive(dentry) &&
1104 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1106 } else if (dentry_lease_is_valid(dentry) ||
1107 dir_lease_is_valid(dir, dentry)) {
1108 if (d_really_is_positive(dentry))
1109 valid = ceph_is_any_caps(d_inode(dentry));
1115 struct ceph_mds_client *mdsc =
1116 ceph_sb_to_client(dir->i_sb)->mdsc;
1117 struct ceph_mds_request *req;
1120 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1121 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1122 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1124 req->r_dentry = dget(dentry);
1125 req->r_num_caps = 2;
1127 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1128 if (ceph_security_xattr_wanted(dir))
1129 mask |= CEPH_CAP_XATTR_SHARED;
1130 req->r_args.getattr.mask = mask;
1132 req->r_locked_dir = dir;
1133 err = ceph_mdsc_do_request(mdsc, NULL, req);
1134 if (err == 0 || err == -ENOENT) {
1135 if (dentry == req->r_dentry) {
1136 valid = !d_unhashed(dentry);
1138 d_invalidate(req->r_dentry);
1142 ceph_mdsc_put_request(req);
1143 dout("d_revalidate %p lookup result=%d\n",
1148 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1150 ceph_dentry_lru_touch(dentry);
1152 ceph_dir_clear_complete(dir);
1160 * Release our ceph_dentry_info.
1162 static void ceph_d_release(struct dentry *dentry)
1164 struct ceph_dentry_info *di = ceph_dentry(dentry);
1166 dout("d_release %p\n", dentry);
1167 ceph_dentry_lru_del(dentry);
1168 if (di->lease_session)
1169 ceph_put_mds_session(di->lease_session);
1170 kmem_cache_free(ceph_dentry_cachep, di);
1171 dentry->d_fsdata = NULL;
1174 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1178 * Eventually, we'll want to revalidate snapped metadata
1179 * too... probably...
1185 * When the VFS prunes a dentry from the cache, we need to clear the
1186 * complete flag on the parent directory.
1188 * Called under dentry->d_lock.
1190 static void ceph_d_prune(struct dentry *dentry)
1192 dout("ceph_d_prune %p\n", dentry);
1194 /* do we have a valid parent? */
1195 if (IS_ROOT(dentry))
1198 /* if we are not hashed, we don't affect dir's completeness */
1199 if (d_unhashed(dentry))
1203 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1204 * cleared until d_release
1206 ceph_dir_clear_complete(d_inode(dentry->d_parent));
1210 * read() on a dir. This weird interface hack only works if mounted
1211 * with '-o dirstat'.
1213 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1216 struct ceph_file_info *cf = file->private_data;
1217 struct inode *inode = file_inode(file);
1218 struct ceph_inode_info *ci = ceph_inode(inode);
1220 const int bufsize = 1024;
1222 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1225 if (!cf->dir_info) {
1226 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1230 snprintf(cf->dir_info, bufsize,
1233 " subdirs: %20lld\n"
1234 "rentries: %20lld\n"
1236 " rsubdirs: %20lld\n"
1238 "rctime: %10ld.%09ld\n",
1239 ci->i_files + ci->i_subdirs,
1242 ci->i_rfiles + ci->i_rsubdirs,
1246 (long)ci->i_rctime.tv_sec,
1247 (long)ci->i_rctime.tv_nsec);
1250 if (*ppos >= cf->dir_info_len)
1252 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1253 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1256 *ppos += (size - left);
1261 * We maintain a private dentry LRU.
1263 * FIXME: this needs to be changed to a per-mds lru to be useful.
1265 void ceph_dentry_lru_add(struct dentry *dn)
1267 struct ceph_dentry_info *di = ceph_dentry(dn);
1268 struct ceph_mds_client *mdsc;
1270 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1271 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1272 spin_lock(&mdsc->dentry_lru_lock);
1273 list_add_tail(&di->lru, &mdsc->dentry_lru);
1275 spin_unlock(&mdsc->dentry_lru_lock);
1278 void ceph_dentry_lru_touch(struct dentry *dn)
1280 struct ceph_dentry_info *di = ceph_dentry(dn);
1281 struct ceph_mds_client *mdsc;
1283 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1285 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1286 spin_lock(&mdsc->dentry_lru_lock);
1287 list_move_tail(&di->lru, &mdsc->dentry_lru);
1288 spin_unlock(&mdsc->dentry_lru_lock);
1291 void ceph_dentry_lru_del(struct dentry *dn)
1293 struct ceph_dentry_info *di = ceph_dentry(dn);
1294 struct ceph_mds_client *mdsc;
1296 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1297 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1298 spin_lock(&mdsc->dentry_lru_lock);
1299 list_del_init(&di->lru);
1301 spin_unlock(&mdsc->dentry_lru_lock);
1305 * Return name hash for a given dentry. This is dependent on
1306 * the parent directory's hash function.
1308 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1310 struct ceph_inode_info *dci = ceph_inode(dir);
1312 switch (dci->i_dir_layout.dl_dir_hash) {
1313 case 0: /* for backward compat */
1314 case CEPH_STR_HASH_LINUX:
1315 return dn->d_name.hash;
1318 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1319 dn->d_name.name, dn->d_name.len);
1323 const struct file_operations ceph_dir_fops = {
1324 .read = ceph_read_dir,
1325 .iterate = ceph_readdir,
1326 .llseek = ceph_dir_llseek,
1328 .release = ceph_release,
1329 .unlocked_ioctl = ceph_ioctl,
1330 .fsync = ceph_fsync,
1333 const struct file_operations ceph_snapdir_fops = {
1334 .iterate = ceph_readdir,
1335 .llseek = ceph_dir_llseek,
1337 .release = ceph_release,
1340 const struct inode_operations ceph_dir_iops = {
1341 .lookup = ceph_lookup,
1342 .permission = ceph_permission,
1343 .getattr = ceph_getattr,
1344 .setattr = ceph_setattr,
1345 .setxattr = ceph_setxattr,
1346 .getxattr = ceph_getxattr,
1347 .listxattr = ceph_listxattr,
1348 .removexattr = ceph_removexattr,
1349 .get_acl = ceph_get_acl,
1350 .set_acl = ceph_set_acl,
1351 .mknod = ceph_mknod,
1352 .symlink = ceph_symlink,
1353 .mkdir = ceph_mkdir,
1355 .unlink = ceph_unlink,
1356 .rmdir = ceph_unlink,
1357 .rename = ceph_rename,
1358 .create = ceph_create,
1359 .atomic_open = ceph_atomic_open,
1362 const struct inode_operations ceph_snapdir_iops = {
1363 .lookup = ceph_lookup,
1364 .permission = ceph_permission,
1365 .getattr = ceph_getattr,
1366 .mkdir = ceph_mkdir,
1367 .rmdir = ceph_unlink,
1368 .rename = ceph_rename,
1371 const struct dentry_operations ceph_dentry_ops = {
1372 .d_revalidate = ceph_d_revalidate,
1373 .d_release = ceph_d_release,
1374 .d_prune = ceph_d_prune,
1377 const struct dentry_operations ceph_snapdir_dentry_ops = {
1378 .d_revalidate = ceph_snapdir_d_revalidate,
1379 .d_release = ceph_d_release,
1382 const struct dentry_operations ceph_snap_dentry_ops = {
1383 .d_release = ceph_d_release,
1384 .d_prune = ceph_d_prune,