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>
8 #include <linux/xattr.h>
11 #include "mds_client.h"
14 * Directory operations: readdir, lookup, create, link, unlink,
19 * Ceph MDS operations are specified in terms of a base ino and
20 * relative path. Thus, the client can specify an operation on a
21 * specific inode (e.g., a getattr due to fstat(2)), or as a path
22 * relative to, say, the root directory.
24 * Normally, we limit ourselves to strict inode ops (no path component)
25 * or dentry operations (a single path component relative to an ino). The
26 * exception to this is open_root_dentry(), which will open the mount
30 const 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 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
44 return -ENOMEM; /* oh well */
46 spin_lock(&dentry->d_lock);
47 if (dentry->d_fsdata) {
49 kmem_cache_free(ceph_dentry_cachep, di);
53 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP)
54 d_set_d_op(dentry, &ceph_dentry_ops);
55 else if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
56 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
58 d_set_d_op(dentry, &ceph_snap_dentry_ops);
61 di->lease_session = NULL;
62 dentry->d_time = jiffies;
63 /* avoid reordering d_fsdata setup so that the check above is safe */
65 dentry->d_fsdata = di;
66 ceph_dentry_lru_add(dentry);
68 spin_unlock(&dentry->d_lock);
73 * for readdir, we encode the directory frag and offset within that
76 static unsigned fpos_frag(loff_t p)
80 static unsigned fpos_off(loff_t p)
82 return p & 0xffffffff;
85 static int fpos_cmp(loff_t l, loff_t r)
87 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
90 return (int)(fpos_off(l) - fpos_off(r));
94 * make note of the last dentry we read, so we can
95 * continue at the same lexicographical point,
96 * regardless of what dir changes take place on the
99 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
100 int len, unsigned next_offset)
102 char *buf = kmalloc(len+1, GFP_KERNEL);
105 kfree(fi->last_name);
107 memcpy(fi->last_name, name, len);
108 fi->last_name[len] = 0;
109 fi->next_offset = next_offset;
110 dout("note_last_dentry '%s'\n", fi->last_name);
115 * When possible, we try to satisfy a readdir by peeking at the
116 * dcache. We make this work by carefully ordering dentries on
117 * d_child when we initially get results back from the MDS, and
118 * falling back to a "normal" sync readdir if any dentries in the dir
121 * Complete dir indicates that we have all dentries in the dir. It is
122 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
123 * the MDS if/when the directory is modified).
125 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
128 struct ceph_file_info *fi = file->private_data;
129 struct dentry *parent = file->f_path.dentry;
130 struct inode *dir = d_inode(parent);
131 struct dentry *dentry, *last = NULL;
132 struct ceph_dentry_info *di;
133 unsigned nsize = PAGE_SIZE / sizeof(struct dentry *);
136 struct ceph_readdir_cache_control cache_ctl = {};
138 dout("__dcache_readdir %p v%u at %llu\n", dir, shared_gen, ctx->pos);
140 /* we can calculate cache index for the first dirfrag */
141 if (ceph_frag_is_leftmost(fpos_frag(ctx->pos))) {
142 cache_ctl.index = fpos_off(ctx->pos) - 2;
143 BUG_ON(cache_ctl.index < 0);
144 ptr_pos = cache_ctl.index * sizeof(struct dentry *);
151 if (ptr_pos >= i_size_read(dir)) {
152 fi->flags |= CEPH_F_ATEND;
158 pgoff = ptr_pos >> PAGE_SHIFT;
159 if (!cache_ctl.page || pgoff != page_index(cache_ctl.page)) {
160 ceph_readdir_cache_release(&cache_ctl);
161 cache_ctl.page = find_lock_page(&dir->i_data, pgoff);
162 if (!cache_ctl.page) {
163 dout(" page %lu not found\n", pgoff);
166 /* reading/filling the cache are serialized by
167 * i_mutex, no need to use page lock */
168 unlock_page(cache_ctl.page);
169 cache_ctl.dentries = kmap(cache_ctl.page);
173 spin_lock(&parent->d_lock);
174 /* check i_size again here, because empty directory can be
175 * marked as complete while not holding the i_mutex. */
176 if (ceph_dir_is_complete_ordered(dir) &&
177 ptr_pos < i_size_read(dir))
178 dentry = cache_ctl.dentries[cache_ctl.index % nsize];
181 spin_unlock(&parent->d_lock);
182 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
189 di = ceph_dentry(dentry);
190 spin_lock(&dentry->d_lock);
191 if (di->lease_shared_gen == shared_gen &&
192 d_really_is_positive(dentry) &&
193 ceph_snap(d_inode(dentry)) != CEPH_SNAPDIR &&
194 ceph_ino(d_inode(dentry)) != CEPH_INO_CEPH &&
195 fpos_cmp(ctx->pos, di->offset) <= 0) {
198 spin_unlock(&dentry->d_lock);
201 dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
202 dentry, dentry, d_inode(dentry));
203 ctx->pos = di->offset;
204 if (!dir_emit(ctx, dentry->d_name.name,
206 ceph_translate_ino(dentry->d_sb,
207 d_inode(dentry)->i_ino),
208 d_inode(dentry)->i_mode >> 12)) {
223 ptr_pos += sizeof(struct dentry *);
225 ceph_readdir_cache_release(&cache_ctl);
228 di = ceph_dentry(last);
229 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
230 fpos_off(di->offset) + 1);
238 static int ceph_readdir(struct file *file, struct dir_context *ctx)
240 struct ceph_file_info *fi = file->private_data;
241 struct inode *inode = file_inode(file);
242 struct ceph_inode_info *ci = ceph_inode(inode);
243 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
244 struct ceph_mds_client *mdsc = fsc->mdsc;
245 unsigned frag = fpos_frag(ctx->pos);
246 int off = fpos_off(ctx->pos);
249 struct ceph_mds_reply_info_parsed *rinfo;
251 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
252 if (fi->flags & CEPH_F_ATEND)
255 /* always start with . and .. */
257 dout("readdir off 0 -> '.'\n");
258 if (!dir_emit(ctx, ".", 1,
259 ceph_translate_ino(inode->i_sb, inode->i_ino),
260 inode->i_mode >> 12))
266 ino_t ino = parent_ino(file->f_path.dentry);
267 dout("readdir off 1 -> '..'\n");
268 if (!dir_emit(ctx, "..", 2,
269 ceph_translate_ino(inode->i_sb, ino),
270 inode->i_mode >> 12))
276 /* can we use the dcache? */
277 spin_lock(&ci->i_ceph_lock);
278 if (ceph_test_mount_opt(fsc, DCACHE) &&
279 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
280 ceph_snap(inode) != CEPH_SNAPDIR &&
281 __ceph_dir_is_complete_ordered(ci) &&
282 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
283 u32 shared_gen = ci->i_shared_gen;
284 spin_unlock(&ci->i_ceph_lock);
285 err = __dcache_readdir(file, ctx, shared_gen);
288 frag = fpos_frag(ctx->pos);
289 off = fpos_off(ctx->pos);
291 spin_unlock(&ci->i_ceph_lock);
294 /* proceed with a normal readdir */
296 /* do we have the correct frag content buffered? */
297 if (fi->frag != frag || fi->last_readdir == NULL) {
298 struct ceph_mds_request *req;
299 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
300 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
302 /* discard old result, if any */
303 if (fi->last_readdir) {
304 ceph_mdsc_put_request(fi->last_readdir);
305 fi->last_readdir = NULL;
308 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
309 ceph_vinop(inode), frag, fi->last_name);
310 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
313 err = ceph_alloc_readdir_reply_buffer(req, inode);
315 ceph_mdsc_put_request(req);
318 /* hints to request -> mds selection code */
319 req->r_direct_mode = USE_AUTH_MDS;
320 req->r_direct_hash = ceph_frag_value(frag);
321 req->r_direct_is_hash = true;
323 req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
325 ceph_mdsc_put_request(req);
329 req->r_dir_release_cnt = fi->dir_release_count;
330 req->r_dir_ordered_cnt = fi->dir_ordered_count;
331 req->r_readdir_cache_idx = fi->readdir_cache_idx;
332 req->r_readdir_offset = fi->next_offset;
333 req->r_args.readdir.frag = cpu_to_le32(frag);
335 req->r_inode = inode;
337 req->r_dentry = dget(file->f_path.dentry);
338 err = ceph_mdsc_do_request(mdsc, NULL, req);
340 ceph_mdsc_put_request(req);
343 dout("readdir got and parsed readdir result=%d"
344 " on frag %x, end=%d, complete=%d\n", err, frag,
345 (int)req->r_reply_info.dir_end,
346 (int)req->r_reply_info.dir_complete);
349 /* note next offset and last dentry name */
350 rinfo = &req->r_reply_info;
351 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
352 frag = le32_to_cpu(rinfo->dir_dir->frag);
353 off = req->r_readdir_offset;
354 fi->next_offset = off;
358 fi->offset = fi->next_offset;
359 fi->last_readdir = req;
361 if (req->r_did_prepopulate) {
362 fi->readdir_cache_idx = req->r_readdir_cache_idx;
363 if (fi->readdir_cache_idx < 0) {
364 /* preclude from marking dir ordered */
365 fi->dir_ordered_count = 0;
366 } else if (ceph_frag_is_leftmost(frag) && off == 2) {
367 /* note dir version at start of readdir so
368 * we can tell if any dentries get dropped */
369 fi->dir_release_count = req->r_dir_release_cnt;
370 fi->dir_ordered_count = req->r_dir_ordered_cnt;
373 dout("readdir !did_prepopulate");
374 /* disable readdir cache */
375 fi->readdir_cache_idx = -1;
376 /* preclude from marking dir complete */
377 fi->dir_release_count = 0;
380 if (req->r_reply_info.dir_end) {
381 kfree(fi->last_name);
382 fi->last_name = NULL;
383 if (ceph_frag_is_rightmost(frag))
388 err = note_last_dentry(fi,
389 rinfo->dir_dname[rinfo->dir_nr-1],
390 rinfo->dir_dname_len[rinfo->dir_nr-1],
391 fi->next_offset + rinfo->dir_nr);
397 rinfo = &fi->last_readdir->r_reply_info;
398 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
399 rinfo->dir_nr, off, fi->offset);
401 ctx->pos = ceph_make_fpos(frag, off);
402 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
403 struct ceph_mds_reply_inode *in =
404 rinfo->dir_in[off - fi->offset].in;
405 struct ceph_vino vino;
408 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
409 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
410 rinfo->dir_dname_len[off - fi->offset],
411 rinfo->dir_dname[off - fi->offset], in);
413 ftype = le32_to_cpu(in->mode) >> 12;
414 vino.ino = le64_to_cpu(in->ino);
415 vino.snap = le64_to_cpu(in->snapid);
416 ino = ceph_vino_to_ino(vino);
418 rinfo->dir_dname[off - fi->offset],
419 rinfo->dir_dname_len[off - fi->offset],
420 ceph_translate_ino(inode->i_sb, ino), ftype)) {
421 dout("filldir stopping us...\n");
429 ceph_mdsc_put_request(fi->last_readdir);
430 fi->last_readdir = NULL;
435 if (!ceph_frag_is_rightmost(frag)) {
436 frag = ceph_frag_next(frag);
438 ctx->pos = ceph_make_fpos(frag, off);
439 dout("readdir next frag is %x\n", frag);
442 fi->flags |= CEPH_F_ATEND;
445 * if dir_release_count still matches the dir, no dentries
446 * were released during the whole readdir, and we should have
447 * the complete dir contents in our cache.
449 if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
450 spin_lock(&ci->i_ceph_lock);
451 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
452 dout(" marking %p complete and ordered\n", inode);
453 /* use i_size to track number of entries in
455 BUG_ON(fi->readdir_cache_idx < 0);
456 i_size_write(inode, fi->readdir_cache_idx *
457 sizeof(struct dentry*));
459 dout(" marking %p complete\n", inode);
461 __ceph_dir_set_complete(ci, fi->dir_release_count,
462 fi->dir_ordered_count);
463 spin_unlock(&ci->i_ceph_lock);
466 dout("readdir %p file %p done.\n", inode, file);
470 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
472 if (fi->last_readdir) {
473 ceph_mdsc_put_request(fi->last_readdir);
474 fi->last_readdir = NULL;
476 kfree(fi->last_name);
477 fi->last_name = NULL;
478 fi->dir_release_count = 0;
479 fi->readdir_cache_idx = -1;
480 if (ceph_frag_is_leftmost(frag))
481 fi->next_offset = 2; /* compensate for . and .. */
484 fi->flags &= ~CEPH_F_ATEND;
487 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
489 struct ceph_file_info *fi = file->private_data;
490 struct inode *inode = file->f_mapping->host;
491 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
498 offset += file->f_pos;
502 retval = -EOPNOTSUPP;
508 if (offset != file->f_pos) {
509 file->f_pos = offset;
511 fi->flags &= ~CEPH_F_ATEND;
516 fpos_frag(offset) != fi->frag ||
517 fpos_off(offset) < fi->offset) {
518 /* discard buffered readdir content on seekdir(0), or
519 * seek to new frag, or seek prior to current chunk */
520 dout("dir_llseek dropping %p content\n", file);
521 reset_readdir(fi, fpos_frag(offset));
522 } else if (fpos_cmp(offset, old_offset) > 0) {
523 /* reset dir_release_count if we did a forward seek */
524 fi->dir_release_count = 0;
525 fi->readdir_cache_idx = -1;
534 * Handle lookups for the hidden .snap directory.
536 int ceph_handle_snapdir(struct ceph_mds_request *req,
537 struct dentry *dentry, int err)
539 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
540 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
543 if (err == -ENOENT &&
544 ceph_snap(parent) == CEPH_NOSNAP &&
545 strcmp(dentry->d_name.name,
546 fsc->mount_options->snapdir_name) == 0) {
547 struct inode *inode = ceph_get_snapdir(parent);
548 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
549 dentry, dentry, inode);
550 BUG_ON(!d_unhashed(dentry));
551 d_add(dentry, inode);
558 * Figure out final result of a lookup/open request.
560 * Mainly, make sure we return the final req->r_dentry (if it already
561 * existed) in place of the original VFS-provided dentry when they
564 * Gracefully handle the case where the MDS replies with -ENOENT and
565 * no trace (which it may do, at its discretion, e.g., if it doesn't
566 * care to issue a lease on the negative dentry).
568 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
569 struct dentry *dentry, int err)
571 if (err == -ENOENT) {
574 if (!req->r_reply_info.head->is_dentry) {
575 dout("ENOENT and no trace, dentry %p inode %p\n",
576 dentry, d_inode(dentry));
577 if (d_really_is_positive(dentry)) {
586 dentry = ERR_PTR(err);
587 else if (dentry != req->r_dentry)
588 dentry = dget(req->r_dentry); /* we got spliced */
594 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
596 return ceph_ino(inode) == CEPH_INO_ROOT &&
597 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
601 * Look up a single dir entry. If there is a lookup intent, inform
602 * the MDS so that it gets our 'caps wanted' value in a single op.
604 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
607 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
608 struct ceph_mds_client *mdsc = fsc->mdsc;
609 struct ceph_mds_request *req;
614 dout("lookup %p dentry %p '%pd'\n",
615 dir, dentry, dentry);
617 if (dentry->d_name.len > NAME_MAX)
618 return ERR_PTR(-ENAMETOOLONG);
620 err = ceph_init_dentry(dentry);
624 /* can we conclude ENOENT locally? */
625 if (d_really_is_negative(dentry)) {
626 struct ceph_inode_info *ci = ceph_inode(dir);
627 struct ceph_dentry_info *di = ceph_dentry(dentry);
629 spin_lock(&ci->i_ceph_lock);
630 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
631 if (strncmp(dentry->d_name.name,
632 fsc->mount_options->snapdir_name,
633 dentry->d_name.len) &&
634 !is_root_ceph_dentry(dir, dentry) &&
635 ceph_test_mount_opt(fsc, DCACHE) &&
636 __ceph_dir_is_complete(ci) &&
637 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
638 spin_unlock(&ci->i_ceph_lock);
639 dout(" dir %p complete, -ENOENT\n", dir);
641 di->lease_shared_gen = ci->i_shared_gen;
644 spin_unlock(&ci->i_ceph_lock);
647 op = ceph_snap(dir) == CEPH_SNAPDIR ?
648 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
649 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
651 return ERR_CAST(req);
652 req->r_dentry = dget(dentry);
655 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
656 if (ceph_security_xattr_wanted(dir))
657 mask |= CEPH_CAP_XATTR_SHARED;
658 req->r_args.getattr.mask = cpu_to_le32(mask);
660 req->r_locked_dir = dir;
661 err = ceph_mdsc_do_request(mdsc, NULL, req);
662 err = ceph_handle_snapdir(req, dentry, err);
663 dentry = ceph_finish_lookup(req, dentry, err);
664 ceph_mdsc_put_request(req); /* will dput(dentry) */
665 dout("lookup result=%p\n", dentry);
670 * If we do a create but get no trace back from the MDS, follow up with
671 * a lookup (the VFS expects us to link up the provided dentry).
673 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
675 struct dentry *result = ceph_lookup(dir, dentry, 0);
677 if (result && !IS_ERR(result)) {
679 * We created the item, then did a lookup, and found
680 * it was already linked to another inode we already
681 * had in our cache (and thus got spliced). To not
682 * confuse VFS (especially when inode is a directory),
683 * we don't link our dentry to that inode, return an
686 * This event should be rare and it happens only when
687 * we talk to old MDS. Recent MDS does not send traceless
688 * reply for request that creates new inode.
693 return PTR_ERR(result);
696 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
697 umode_t mode, dev_t rdev)
699 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
700 struct ceph_mds_client *mdsc = fsc->mdsc;
701 struct ceph_mds_request *req;
702 struct ceph_acls_info acls = {};
705 if (ceph_snap(dir) != CEPH_NOSNAP)
708 err = ceph_pre_init_acls(dir, &mode, &acls);
712 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
713 dir, dentry, mode, rdev);
714 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
719 req->r_dentry = dget(dentry);
721 req->r_locked_dir = dir;
722 req->r_args.mknod.mode = cpu_to_le32(mode);
723 req->r_args.mknod.rdev = cpu_to_le32(rdev);
724 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
725 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
727 req->r_pagelist = acls.pagelist;
728 acls.pagelist = NULL;
730 err = ceph_mdsc_do_request(mdsc, dir, req);
731 if (!err && !req->r_reply_info.head->is_dentry)
732 err = ceph_handle_notrace_create(dir, dentry);
733 ceph_mdsc_put_request(req);
736 ceph_init_inode_acls(d_inode(dentry), &acls);
739 ceph_release_acls_info(&acls);
743 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
746 return ceph_mknod(dir, dentry, mode, 0);
749 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
752 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
753 struct ceph_mds_client *mdsc = fsc->mdsc;
754 struct ceph_mds_request *req;
757 if (ceph_snap(dir) != CEPH_NOSNAP)
760 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
761 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
766 req->r_path2 = kstrdup(dest, GFP_KERNEL);
769 ceph_mdsc_put_request(req);
772 req->r_locked_dir = dir;
773 req->r_dentry = dget(dentry);
775 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
776 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
777 err = ceph_mdsc_do_request(mdsc, dir, req);
778 if (!err && !req->r_reply_info.head->is_dentry)
779 err = ceph_handle_notrace_create(dir, dentry);
780 ceph_mdsc_put_request(req);
787 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
789 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
790 struct ceph_mds_client *mdsc = fsc->mdsc;
791 struct ceph_mds_request *req;
792 struct ceph_acls_info acls = {};
796 if (ceph_snap(dir) == CEPH_SNAPDIR) {
797 /* mkdir .snap/foo is a MKSNAP */
798 op = CEPH_MDS_OP_MKSNAP;
799 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
801 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
802 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
803 op = CEPH_MDS_OP_MKDIR;
809 err = ceph_pre_init_acls(dir, &mode, &acls);
813 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
819 req->r_dentry = dget(dentry);
821 req->r_locked_dir = dir;
822 req->r_args.mkdir.mode = cpu_to_le32(mode);
823 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
824 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
826 req->r_pagelist = acls.pagelist;
827 acls.pagelist = NULL;
829 err = ceph_mdsc_do_request(mdsc, dir, req);
831 !req->r_reply_info.head->is_target &&
832 !req->r_reply_info.head->is_dentry)
833 err = ceph_handle_notrace_create(dir, dentry);
834 ceph_mdsc_put_request(req);
837 ceph_init_inode_acls(d_inode(dentry), &acls);
840 ceph_release_acls_info(&acls);
844 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
845 struct dentry *dentry)
847 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
848 struct ceph_mds_client *mdsc = fsc->mdsc;
849 struct ceph_mds_request *req;
852 if (ceph_snap(dir) != CEPH_NOSNAP)
855 dout("link in dir %p old_dentry %p dentry %p\n", dir,
857 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
862 req->r_dentry = dget(dentry);
864 req->r_old_dentry = dget(old_dentry);
865 req->r_locked_dir = dir;
866 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
867 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
868 /* release LINK_SHARED on source inode (mds will lock it) */
869 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
870 err = ceph_mdsc_do_request(mdsc, dir, req);
873 } else if (!req->r_reply_info.head->is_dentry) {
874 ihold(d_inode(old_dentry));
875 d_instantiate(dentry, d_inode(old_dentry));
877 ceph_mdsc_put_request(req);
882 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
883 * looks like the link count will hit 0, drop any other caps (other
884 * than PIN) we don't specifically want (due to the file still being
887 static int drop_caps_for_unlink(struct inode *inode)
889 struct ceph_inode_info *ci = ceph_inode(inode);
890 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
892 spin_lock(&ci->i_ceph_lock);
893 if (inode->i_nlink == 1) {
894 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
895 ci->i_ceph_flags |= CEPH_I_NODELAY;
897 spin_unlock(&ci->i_ceph_lock);
902 * rmdir and unlink are differ only by the metadata op code
904 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
906 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
907 struct ceph_mds_client *mdsc = fsc->mdsc;
908 struct inode *inode = d_inode(dentry);
909 struct ceph_mds_request *req;
913 if (ceph_snap(dir) == CEPH_SNAPDIR) {
914 /* rmdir .snap/foo is RMSNAP */
915 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
916 op = CEPH_MDS_OP_RMSNAP;
917 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
918 dout("unlink/rmdir dir %p dn %p inode %p\n",
920 op = d_is_dir(dentry) ?
921 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
924 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
929 req->r_dentry = dget(dentry);
931 req->r_locked_dir = dir;
932 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
933 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
934 req->r_inode_drop = drop_caps_for_unlink(inode);
935 err = ceph_mdsc_do_request(mdsc, dir, req);
936 if (!err && !req->r_reply_info.head->is_dentry)
938 ceph_mdsc_put_request(req);
943 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
944 struct inode *new_dir, struct dentry *new_dentry)
946 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
947 struct ceph_mds_client *mdsc = fsc->mdsc;
948 struct ceph_mds_request *req;
949 int op = CEPH_MDS_OP_RENAME;
952 if (ceph_snap(old_dir) != ceph_snap(new_dir))
954 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
955 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
956 op = CEPH_MDS_OP_RENAMESNAP;
960 dout("rename dir %p dentry %p to dir %p dentry %p\n",
961 old_dir, old_dentry, new_dir, new_dentry);
962 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
966 req->r_dentry = dget(new_dentry);
968 req->r_old_dentry = dget(old_dentry);
969 req->r_old_dentry_dir = old_dir;
970 req->r_locked_dir = new_dir;
971 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
972 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
973 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
974 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
975 /* release LINK_RDCACHE on source inode (mds will lock it) */
976 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
977 if (d_really_is_positive(new_dentry))
978 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
979 err = ceph_mdsc_do_request(mdsc, old_dir, req);
980 if (!err && !req->r_reply_info.head->is_dentry) {
982 * Normally d_move() is done by fill_trace (called by
983 * do_request, above). If there is no trace, we need
987 /* d_move screws up sibling dentries' offsets */
988 ceph_dir_clear_complete(old_dir);
989 ceph_dir_clear_complete(new_dir);
991 d_move(old_dentry, new_dentry);
993 /* ensure target dentry is invalidated, despite
994 rehashing bug in vfs_rename_dir */
995 ceph_invalidate_dentry_lease(new_dentry);
997 ceph_mdsc_put_request(req);
1002 * Ensure a dentry lease will no longer revalidate.
1004 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1006 spin_lock(&dentry->d_lock);
1007 dentry->d_time = jiffies;
1008 ceph_dentry(dentry)->lease_shared_gen = 0;
1009 spin_unlock(&dentry->d_lock);
1013 * Check if dentry lease is valid. If not, delete the lease. Try to
1014 * renew if the least is more than half up.
1016 static int dentry_lease_is_valid(struct dentry *dentry)
1018 struct ceph_dentry_info *di;
1019 struct ceph_mds_session *s;
1023 struct ceph_mds_session *session = NULL;
1024 struct inode *dir = NULL;
1027 spin_lock(&dentry->d_lock);
1028 di = ceph_dentry(dentry);
1029 if (di->lease_session) {
1030 s = di->lease_session;
1031 spin_lock(&s->s_gen_ttl_lock);
1034 spin_unlock(&s->s_gen_ttl_lock);
1036 if (di->lease_gen == gen &&
1037 time_before(jiffies, dentry->d_time) &&
1038 time_before(jiffies, ttl)) {
1040 if (di->lease_renew_after &&
1041 time_after(jiffies, di->lease_renew_after)) {
1042 /* we should renew */
1043 dir = d_inode(dentry->d_parent);
1044 session = ceph_get_mds_session(s);
1045 seq = di->lease_seq;
1046 di->lease_renew_after = 0;
1047 di->lease_renew_from = jiffies;
1051 spin_unlock(&dentry->d_lock);
1054 ceph_mdsc_lease_send_msg(session, dir, dentry,
1055 CEPH_MDS_LEASE_RENEW, seq);
1056 ceph_put_mds_session(session);
1058 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1063 * Check if directory-wide content lease/cap is valid.
1065 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1067 struct ceph_inode_info *ci = ceph_inode(dir);
1068 struct ceph_dentry_info *di = ceph_dentry(dentry);
1071 spin_lock(&ci->i_ceph_lock);
1072 if (ci->i_shared_gen == di->lease_shared_gen)
1073 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1074 spin_unlock(&ci->i_ceph_lock);
1075 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1076 dir, (unsigned)ci->i_shared_gen, dentry,
1077 (unsigned)di->lease_shared_gen, valid);
1082 * Check if cached dentry can be trusted.
1084 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1087 struct dentry *parent;
1090 if (flags & LOOKUP_RCU)
1093 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1094 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1096 parent = dget_parent(dentry);
1097 dir = d_inode(parent);
1099 /* always trust cached snapped dentries, snapdir dentry */
1100 if (ceph_snap(dir) != CEPH_NOSNAP) {
1101 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1102 dentry, d_inode(dentry));
1104 } else if (d_really_is_positive(dentry) &&
1105 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1107 } else if (dentry_lease_is_valid(dentry) ||
1108 dir_lease_is_valid(dir, dentry)) {
1109 if (d_really_is_positive(dentry))
1110 valid = ceph_is_any_caps(d_inode(dentry));
1116 struct ceph_mds_client *mdsc =
1117 ceph_sb_to_client(dir->i_sb)->mdsc;
1118 struct ceph_mds_request *req;
1121 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1122 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1123 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1125 req->r_dentry = dget(dentry);
1126 req->r_num_caps = 2;
1128 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1129 if (ceph_security_xattr_wanted(dir))
1130 mask |= CEPH_CAP_XATTR_SHARED;
1131 req->r_args.getattr.mask = mask;
1133 req->r_locked_dir = dir;
1134 err = ceph_mdsc_do_request(mdsc, NULL, req);
1135 if (err == 0 || err == -ENOENT) {
1136 if (dentry == req->r_dentry) {
1137 valid = !d_unhashed(dentry);
1139 d_invalidate(req->r_dentry);
1143 ceph_mdsc_put_request(req);
1144 dout("d_revalidate %p lookup result=%d\n",
1149 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1151 ceph_dentry_lru_touch(dentry);
1153 ceph_dir_clear_complete(dir);
1161 * Release our ceph_dentry_info.
1163 static void ceph_d_release(struct dentry *dentry)
1165 struct ceph_dentry_info *di = ceph_dentry(dentry);
1167 dout("d_release %p\n", dentry);
1168 ceph_dentry_lru_del(dentry);
1169 if (di->lease_session)
1170 ceph_put_mds_session(di->lease_session);
1171 kmem_cache_free(ceph_dentry_cachep, di);
1172 dentry->d_fsdata = NULL;
1175 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1179 * Eventually, we'll want to revalidate snapped metadata
1180 * too... probably...
1186 * When the VFS prunes a dentry from the cache, we need to clear the
1187 * complete flag on the parent directory.
1189 * Called under dentry->d_lock.
1191 static void ceph_d_prune(struct dentry *dentry)
1193 dout("ceph_d_prune %p\n", dentry);
1195 /* do we have a valid parent? */
1196 if (IS_ROOT(dentry))
1199 /* if we are not hashed, we don't affect dir's completeness */
1200 if (d_unhashed(dentry))
1204 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1205 * cleared until d_release
1207 ceph_dir_clear_complete(d_inode(dentry->d_parent));
1211 * read() on a dir. This weird interface hack only works if mounted
1212 * with '-o dirstat'.
1214 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1217 struct ceph_file_info *cf = file->private_data;
1218 struct inode *inode = file_inode(file);
1219 struct ceph_inode_info *ci = ceph_inode(inode);
1221 const int bufsize = 1024;
1223 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1226 if (!cf->dir_info) {
1227 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1231 snprintf(cf->dir_info, bufsize,
1234 " subdirs: %20lld\n"
1235 "rentries: %20lld\n"
1237 " rsubdirs: %20lld\n"
1239 "rctime: %10ld.%09ld\n",
1240 ci->i_files + ci->i_subdirs,
1243 ci->i_rfiles + ci->i_rsubdirs,
1247 (long)ci->i_rctime.tv_sec,
1248 (long)ci->i_rctime.tv_nsec);
1251 if (*ppos >= cf->dir_info_len)
1253 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1254 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1257 *ppos += (size - left);
1262 * We maintain a private dentry LRU.
1264 * FIXME: this needs to be changed to a per-mds lru to be useful.
1266 void ceph_dentry_lru_add(struct dentry *dn)
1268 struct ceph_dentry_info *di = ceph_dentry(dn);
1269 struct ceph_mds_client *mdsc;
1271 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1272 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1273 spin_lock(&mdsc->dentry_lru_lock);
1274 list_add_tail(&di->lru, &mdsc->dentry_lru);
1276 spin_unlock(&mdsc->dentry_lru_lock);
1279 void ceph_dentry_lru_touch(struct dentry *dn)
1281 struct ceph_dentry_info *di = ceph_dentry(dn);
1282 struct ceph_mds_client *mdsc;
1284 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1286 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1287 spin_lock(&mdsc->dentry_lru_lock);
1288 list_move_tail(&di->lru, &mdsc->dentry_lru);
1289 spin_unlock(&mdsc->dentry_lru_lock);
1292 void ceph_dentry_lru_del(struct dentry *dn)
1294 struct ceph_dentry_info *di = ceph_dentry(dn);
1295 struct ceph_mds_client *mdsc;
1297 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1298 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1299 spin_lock(&mdsc->dentry_lru_lock);
1300 list_del_init(&di->lru);
1302 spin_unlock(&mdsc->dentry_lru_lock);
1306 * Return name hash for a given dentry. This is dependent on
1307 * the parent directory's hash function.
1309 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1311 struct ceph_inode_info *dci = ceph_inode(dir);
1313 switch (dci->i_dir_layout.dl_dir_hash) {
1314 case 0: /* for backward compat */
1315 case CEPH_STR_HASH_LINUX:
1316 return dn->d_name.hash;
1319 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1320 dn->d_name.name, dn->d_name.len);
1324 const struct file_operations ceph_dir_fops = {
1325 .read = ceph_read_dir,
1326 .iterate = ceph_readdir,
1327 .llseek = ceph_dir_llseek,
1329 .release = ceph_release,
1330 .unlocked_ioctl = ceph_ioctl,
1331 .fsync = ceph_fsync,
1334 const struct file_operations ceph_snapdir_fops = {
1335 .iterate = ceph_readdir,
1336 .llseek = ceph_dir_llseek,
1338 .release = ceph_release,
1341 const struct inode_operations ceph_dir_iops = {
1342 .lookup = ceph_lookup,
1343 .permission = ceph_permission,
1344 .getattr = ceph_getattr,
1345 .setattr = ceph_setattr,
1346 .setxattr = generic_setxattr,
1347 .getxattr = generic_getxattr,
1348 .listxattr = ceph_listxattr,
1349 .removexattr = generic_removexattr,
1350 .get_acl = ceph_get_acl,
1351 .set_acl = ceph_set_acl,
1352 .mknod = ceph_mknod,
1353 .symlink = ceph_symlink,
1354 .mkdir = ceph_mkdir,
1356 .unlink = ceph_unlink,
1357 .rmdir = ceph_unlink,
1358 .rename = ceph_rename,
1359 .create = ceph_create,
1360 .atomic_open = ceph_atomic_open,
1363 const struct inode_operations ceph_snapdir_iops = {
1364 .lookup = ceph_lookup,
1365 .permission = ceph_permission,
1366 .getattr = ceph_getattr,
1367 .mkdir = ceph_mkdir,
1368 .rmdir = ceph_unlink,
1369 .rename = ceph_rename,
1372 const struct dentry_operations ceph_dentry_ops = {
1373 .d_revalidate = ceph_d_revalidate,
1374 .d_release = ceph_d_release,
1375 .d_prune = ceph_d_prune,
1378 const struct dentry_operations ceph_snapdir_dentry_ops = {
1379 .d_revalidate = ceph_snapdir_d_revalidate,
1380 .d_release = ceph_d_release,
1383 const struct dentry_operations ceph_snap_dentry_ops = {
1384 .d_release = ceph_d_release,
1385 .d_prune = ceph_d_prune,