offb: Fix bug in calculating requested vram size
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ceph / dir.c
blobef8f08c343e8936df6f6fd2cde5eb895f93d8f83
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>
9 #include "super.h"
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
14 * rename, etc.
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
26 * point by name.
29 const struct inode_operations ceph_dir_iops;
30 const struct file_operations ceph_dir_fops;
31 const struct dentry_operations ceph_dentry_ops;
34 * Initialize ceph dentry state.
36 int ceph_init_dentry(struct dentry *dentry)
38 struct ceph_dentry_info *di;
40 if (dentry->d_fsdata)
41 return 0;
43 if (dentry->d_parent == NULL || /* nfs fh_to_dentry */
44 ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
45 d_set_d_op(dentry, &ceph_dentry_ops);
46 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
47 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
48 else
49 d_set_d_op(dentry, &ceph_snap_dentry_ops);
51 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
52 if (!di)
53 return -ENOMEM; /* oh well */
55 spin_lock(&dentry->d_lock);
56 if (dentry->d_fsdata) {
57 /* lost a race */
58 kmem_cache_free(ceph_dentry_cachep, di);
59 goto out_unlock;
61 di->dentry = dentry;
62 di->lease_session = NULL;
63 dentry->d_fsdata = di;
64 dentry->d_time = jiffies;
65 ceph_dentry_lru_add(dentry);
66 out_unlock:
67 spin_unlock(&dentry->d_lock);
68 return 0;
74 * for readdir, we encode the directory frag and offset within that
75 * frag into f_pos.
77 static unsigned fpos_frag(loff_t p)
79 return p >> 32;
81 static unsigned fpos_off(loff_t p)
83 return p & 0xffffffff;
87 * When possible, we try to satisfy a readdir by peeking at the
88 * dcache. We make this work by carefully ordering dentries on
89 * d_u.d_child when we initially get results back from the MDS, and
90 * falling back to a "normal" sync readdir if any dentries in the dir
91 * are dropped.
93 * I_COMPLETE tells indicates we have all dentries in the dir. It is
94 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
95 * the MDS if/when the directory is modified).
97 static int __dcache_readdir(struct file *filp,
98 void *dirent, filldir_t filldir)
100 struct ceph_file_info *fi = filp->private_data;
101 struct dentry *parent = filp->f_dentry;
102 struct inode *dir = parent->d_inode;
103 struct list_head *p;
104 struct dentry *dentry, *last;
105 struct ceph_dentry_info *di;
106 int err = 0;
108 /* claim ref on last dentry we returned */
109 last = fi->dentry;
110 fi->dentry = NULL;
112 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
113 last);
115 spin_lock(&parent->d_lock);
117 /* start at beginning? */
118 if (filp->f_pos == 2 || last == NULL ||
119 filp->f_pos < ceph_dentry(last)->offset) {
120 if (list_empty(&parent->d_subdirs))
121 goto out_unlock;
122 p = parent->d_subdirs.prev;
123 dout(" initial p %p/%p\n", p->prev, p->next);
124 } else {
125 p = last->d_u.d_child.prev;
128 more:
129 dentry = list_entry(p, struct dentry, d_u.d_child);
130 di = ceph_dentry(dentry);
131 while (1) {
132 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
133 d_unhashed(dentry) ? "!hashed" : "hashed",
134 parent->d_subdirs.prev, parent->d_subdirs.next);
135 if (p == &parent->d_subdirs) {
136 fi->at_end = 1;
137 goto out_unlock;
139 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
140 if (!d_unhashed(dentry) && dentry->d_inode &&
141 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
142 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
143 filp->f_pos <= di->offset)
144 break;
145 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
146 dentry->d_name.len, dentry->d_name.name, di->offset,
147 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
148 !dentry->d_inode ? " null" : "");
149 spin_unlock(&dentry->d_lock);
150 p = p->prev;
151 dentry = list_entry(p, struct dentry, d_u.d_child);
152 di = ceph_dentry(dentry);
155 dget_dlock(dentry);
156 spin_unlock(&dentry->d_lock);
157 spin_unlock(&parent->d_lock);
159 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
160 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
161 filp->f_pos = di->offset;
162 err = filldir(dirent, dentry->d_name.name,
163 dentry->d_name.len, di->offset,
164 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
165 dentry->d_inode->i_mode >> 12);
167 if (last) {
168 if (err < 0) {
169 /* remember our position */
170 fi->dentry = last;
171 fi->next_offset = di->offset;
172 } else {
173 dput(last);
176 last = dentry;
178 if (err < 0)
179 goto out;
181 filp->f_pos++;
183 /* make sure a dentry wasn't dropped while we didn't have parent lock */
184 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
185 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
186 err = -EAGAIN;
187 goto out;
190 spin_lock(&parent->d_lock);
191 p = p->prev; /* advance to next dentry */
192 goto more;
194 out_unlock:
195 spin_unlock(&parent->d_lock);
196 out:
197 if (last)
198 dput(last);
199 return err;
203 * make note of the last dentry we read, so we can
204 * continue at the same lexicographical point,
205 * regardless of what dir changes take place on the
206 * server.
208 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
209 int len)
211 kfree(fi->last_name);
212 fi->last_name = kmalloc(len+1, GFP_NOFS);
213 if (!fi->last_name)
214 return -ENOMEM;
215 memcpy(fi->last_name, name, len);
216 fi->last_name[len] = 0;
217 dout("note_last_dentry '%s'\n", fi->last_name);
218 return 0;
221 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
223 struct ceph_file_info *fi = filp->private_data;
224 struct inode *inode = filp->f_dentry->d_inode;
225 struct ceph_inode_info *ci = ceph_inode(inode);
226 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
227 struct ceph_mds_client *mdsc = fsc->mdsc;
228 unsigned frag = fpos_frag(filp->f_pos);
229 int off = fpos_off(filp->f_pos);
230 int err;
231 u32 ftype;
232 struct ceph_mds_reply_info_parsed *rinfo;
233 const int max_entries = fsc->mount_options->max_readdir;
234 const int max_bytes = fsc->mount_options->max_readdir_bytes;
236 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
237 if (fi->at_end)
238 return 0;
240 /* always start with . and .. */
241 if (filp->f_pos == 0) {
242 /* note dir version at start of readdir so we can tell
243 * if any dentries get dropped */
244 fi->dir_release_count = ci->i_release_count;
246 dout("readdir off 0 -> '.'\n");
247 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
248 ceph_translate_ino(inode->i_sb, inode->i_ino),
249 inode->i_mode >> 12) < 0)
250 return 0;
251 filp->f_pos = 1;
252 off = 1;
254 if (filp->f_pos == 1) {
255 ino_t ino = filp->f_dentry->d_parent->d_inode->i_ino;
256 dout("readdir off 1 -> '..'\n");
257 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
258 ceph_translate_ino(inode->i_sb, ino),
259 inode->i_mode >> 12) < 0)
260 return 0;
261 filp->f_pos = 2;
262 off = 2;
265 /* can we use the dcache? */
266 spin_lock(&inode->i_lock);
267 if ((filp->f_pos == 2 || fi->dentry) &&
268 !ceph_test_mount_opt(fsc, 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 spin_unlock(&inode->i_lock);
273 err = __dcache_readdir(filp, dirent, filldir);
274 if (err != -EAGAIN)
275 return err;
276 } else {
277 spin_unlock(&inode->i_lock);
279 if (fi->dentry) {
280 err = note_last_dentry(fi, fi->dentry->d_name.name,
281 fi->dentry->d_name.len);
282 if (err)
283 return err;
284 dput(fi->dentry);
285 fi->dentry = NULL;
288 /* proceed with a normal readdir */
290 more:
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);
309 if (IS_ERR(req))
310 return PTR_ERR(req);
311 req->r_inode = inode;
312 ihold(inode);
313 req->r_dentry = dget(filp->f_dentry);
314 /* hints to request -> mds selection code */
315 req->r_direct_mode = USE_AUTH_MDS;
316 req->r_direct_hash = ceph_frag_value(frag);
317 req->r_direct_is_hash = true;
318 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
319 req->r_readdir_offset = fi->next_offset;
320 req->r_args.readdir.frag = cpu_to_le32(frag);
321 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
322 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
323 req->r_num_caps = max_entries + 1;
324 err = ceph_mdsc_do_request(mdsc, NULL, req);
325 if (err < 0) {
326 ceph_mdsc_put_request(req);
327 return err;
329 dout("readdir got and parsed readdir result=%d"
330 " on frag %x, end=%d, complete=%d\n", err, frag,
331 (int)req->r_reply_info.dir_end,
332 (int)req->r_reply_info.dir_complete);
334 if (!req->r_did_prepopulate) {
335 dout("readdir !did_prepopulate");
336 fi->dir_release_count--; /* preclude I_COMPLETE */
339 /* note next offset and last dentry name */
340 fi->offset = fi->next_offset;
341 fi->last_readdir = req;
343 if (req->r_reply_info.dir_end) {
344 kfree(fi->last_name);
345 fi->last_name = NULL;
346 if (ceph_frag_is_rightmost(frag))
347 fi->next_offset = 2;
348 else
349 fi->next_offset = 0;
350 } else {
351 rinfo = &req->r_reply_info;
352 err = note_last_dentry(fi,
353 rinfo->dir_dname[rinfo->dir_nr-1],
354 rinfo->dir_dname_len[rinfo->dir_nr-1]);
355 if (err)
356 return err;
357 fi->next_offset += rinfo->dir_nr;
361 rinfo = &fi->last_readdir->r_reply_info;
362 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
363 rinfo->dir_nr, off, fi->offset);
364 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
365 u64 pos = ceph_make_fpos(frag, off);
366 struct ceph_mds_reply_inode *in =
367 rinfo->dir_in[off - fi->offset].in;
368 struct ceph_vino vino;
369 ino_t ino;
371 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
372 off, off - fi->offset, rinfo->dir_nr, pos,
373 rinfo->dir_dname_len[off - fi->offset],
374 rinfo->dir_dname[off - fi->offset], in);
375 BUG_ON(!in);
376 ftype = le32_to_cpu(in->mode) >> 12;
377 vino.ino = le64_to_cpu(in->ino);
378 vino.snap = le64_to_cpu(in->snapid);
379 ino = ceph_vino_to_ino(vino);
380 if (filldir(dirent,
381 rinfo->dir_dname[off - fi->offset],
382 rinfo->dir_dname_len[off - fi->offset],
383 pos,
384 ceph_translate_ino(inode->i_sb, ino), ftype) < 0) {
385 dout("filldir stopping us...\n");
386 return 0;
388 off++;
389 filp->f_pos = pos + 1;
392 if (fi->last_name) {
393 ceph_mdsc_put_request(fi->last_readdir);
394 fi->last_readdir = NULL;
395 goto more;
398 /* more frags? */
399 if (!ceph_frag_is_rightmost(frag)) {
400 frag = ceph_frag_next(frag);
401 off = 0;
402 filp->f_pos = ceph_make_fpos(frag, off);
403 dout("readdir next frag is %x\n", frag);
404 goto more;
406 fi->at_end = 1;
409 * if dir_release_count still matches the dir, no dentries
410 * were released during the whole readdir, and we should have
411 * the complete dir contents in our cache.
413 spin_lock(&inode->i_lock);
414 if (ci->i_release_count == fi->dir_release_count) {
415 dout(" marking %p complete\n", inode);
416 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */
417 ci->i_max_offset = filp->f_pos;
419 spin_unlock(&inode->i_lock);
421 dout("readdir %p filp %p done.\n", inode, filp);
422 return 0;
425 static void reset_readdir(struct ceph_file_info *fi)
427 if (fi->last_readdir) {
428 ceph_mdsc_put_request(fi->last_readdir);
429 fi->last_readdir = NULL;
431 kfree(fi->last_name);
432 fi->last_name = NULL;
433 fi->next_offset = 2; /* compensate for . and .. */
434 if (fi->dentry) {
435 dput(fi->dentry);
436 fi->dentry = NULL;
438 fi->at_end = 0;
441 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
443 struct ceph_file_info *fi = file->private_data;
444 struct inode *inode = file->f_mapping->host;
445 loff_t old_offset = offset;
446 loff_t retval;
448 mutex_lock(&inode->i_mutex);
449 switch (origin) {
450 case SEEK_END:
451 offset += inode->i_size + 2; /* FIXME */
452 break;
453 case SEEK_CUR:
454 offset += file->f_pos;
456 retval = -EINVAL;
457 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
458 if (offset != file->f_pos) {
459 file->f_pos = offset;
460 file->f_version = 0;
461 fi->at_end = 0;
463 retval = offset;
466 * discard buffered readdir content on seekdir(0), or
467 * seek to new frag, or seek prior to current chunk.
469 if (offset == 0 ||
470 fpos_frag(offset) != fpos_frag(old_offset) ||
471 fpos_off(offset) < fi->offset) {
472 dout("dir_llseek dropping %p content\n", file);
473 reset_readdir(fi);
476 /* bump dir_release_count if we did a forward seek */
477 if (offset > old_offset)
478 fi->dir_release_count--;
480 mutex_unlock(&inode->i_mutex);
481 return retval;
485 * Process result of a lookup/open request.
487 * Mainly, make sure we return the final req->r_dentry (if it already
488 * existed) in place of the original VFS-provided dentry when they
489 * differ.
491 * Gracefully handle the case where the MDS replies with -ENOENT and
492 * no trace (which it may do, at its discretion, e.g., if it doesn't
493 * care to issue a lease on the negative dentry).
495 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
496 struct dentry *dentry, int err)
498 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
499 struct inode *parent = dentry->d_parent->d_inode;
501 /* .snap dir? */
502 if (err == -ENOENT &&
503 ceph_snap(parent) == CEPH_NOSNAP &&
504 strcmp(dentry->d_name.name,
505 fsc->mount_options->snapdir_name) == 0) {
506 struct inode *inode = ceph_get_snapdir(parent);
507 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
508 dentry, dentry->d_name.len, dentry->d_name.name, inode);
509 BUG_ON(!d_unhashed(dentry));
510 d_add(dentry, inode);
511 err = 0;
514 if (err == -ENOENT) {
515 /* no trace? */
516 err = 0;
517 if (!req->r_reply_info.head->is_dentry) {
518 dout("ENOENT and no trace, dentry %p inode %p\n",
519 dentry, dentry->d_inode);
520 if (dentry->d_inode) {
521 d_drop(dentry);
522 err = -ENOENT;
523 } else {
524 d_add(dentry, NULL);
528 if (err)
529 dentry = ERR_PTR(err);
530 else if (dentry != req->r_dentry)
531 dentry = dget(req->r_dentry); /* we got spliced */
532 else
533 dentry = NULL;
534 return dentry;
537 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
539 return ceph_ino(inode) == CEPH_INO_ROOT &&
540 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
544 * Look up a single dir entry. If there is a lookup intent, inform
545 * the MDS so that it gets our 'caps wanted' value in a single op.
547 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
548 struct nameidata *nd)
550 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
551 struct ceph_mds_client *mdsc = fsc->mdsc;
552 struct ceph_mds_request *req;
553 int op;
554 int err;
556 dout("lookup %p dentry %p '%.*s'\n",
557 dir, dentry, dentry->d_name.len, dentry->d_name.name);
559 if (dentry->d_name.len > NAME_MAX)
560 return ERR_PTR(-ENAMETOOLONG);
562 err = ceph_init_dentry(dentry);
563 if (err < 0)
564 return ERR_PTR(err);
566 /* open (but not create!) intent? */
567 if (nd &&
568 (nd->flags & LOOKUP_OPEN) &&
569 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
570 !(nd->intent.open.flags & O_CREAT)) {
571 int mode = nd->intent.open.create_mode & ~current->fs->umask;
572 return ceph_lookup_open(dir, dentry, nd, mode, 1);
575 /* can we conclude ENOENT locally? */
576 if (dentry->d_inode == NULL) {
577 struct ceph_inode_info *ci = ceph_inode(dir);
578 struct ceph_dentry_info *di = ceph_dentry(dentry);
580 spin_lock(&dir->i_lock);
581 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
582 if (strncmp(dentry->d_name.name,
583 fsc->mount_options->snapdir_name,
584 dentry->d_name.len) &&
585 !is_root_ceph_dentry(dir, dentry) &&
586 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
587 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
588 spin_unlock(&dir->i_lock);
589 dout(" dir %p complete, -ENOENT\n", dir);
590 d_add(dentry, NULL);
591 di->lease_shared_gen = ci->i_shared_gen;
592 return NULL;
594 spin_unlock(&dir->i_lock);
597 op = ceph_snap(dir) == CEPH_SNAPDIR ?
598 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
599 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
600 if (IS_ERR(req))
601 return ERR_CAST(req);
602 req->r_dentry = dget(dentry);
603 req->r_num_caps = 2;
604 /* we only need inode linkage */
605 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
606 req->r_locked_dir = dir;
607 err = ceph_mdsc_do_request(mdsc, NULL, req);
608 dentry = ceph_finish_lookup(req, dentry, err);
609 ceph_mdsc_put_request(req); /* will dput(dentry) */
610 dout("lookup result=%p\n", dentry);
611 return dentry;
615 * If we do a create but get no trace back from the MDS, follow up with
616 * a lookup (the VFS expects us to link up the provided dentry).
618 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
620 struct dentry *result = ceph_lookup(dir, dentry, NULL);
622 if (result && !IS_ERR(result)) {
624 * We created the item, then did a lookup, and found
625 * it was already linked to another inode we already
626 * had in our cache (and thus got spliced). Link our
627 * dentry to that inode, but don't hash it, just in
628 * case the VFS wants to dereference it.
630 BUG_ON(!result->d_inode);
631 d_instantiate(dentry, result->d_inode);
632 return 0;
634 return PTR_ERR(result);
637 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
638 int mode, dev_t rdev)
640 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
641 struct ceph_mds_client *mdsc = fsc->mdsc;
642 struct ceph_mds_request *req;
643 int err;
645 if (ceph_snap(dir) != CEPH_NOSNAP)
646 return -EROFS;
648 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
649 dir, dentry, mode, rdev);
650 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
651 if (IS_ERR(req)) {
652 d_drop(dentry);
653 return PTR_ERR(req);
655 req->r_dentry = dget(dentry);
656 req->r_num_caps = 2;
657 req->r_locked_dir = dir;
658 req->r_args.mknod.mode = cpu_to_le32(mode);
659 req->r_args.mknod.rdev = cpu_to_le32(rdev);
660 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
661 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
662 err = ceph_mdsc_do_request(mdsc, dir, req);
663 if (!err && !req->r_reply_info.head->is_dentry)
664 err = ceph_handle_notrace_create(dir, dentry);
665 ceph_mdsc_put_request(req);
666 if (err)
667 d_drop(dentry);
668 return err;
671 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
672 struct nameidata *nd)
674 dout("create in dir %p dentry %p name '%.*s'\n",
675 dir, dentry, dentry->d_name.len, dentry->d_name.name);
677 if (ceph_snap(dir) != CEPH_NOSNAP)
678 return -EROFS;
680 if (nd) {
681 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
682 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
683 /* hrm, what should i do here if we get aliased? */
684 if (IS_ERR(dentry))
685 return PTR_ERR(dentry);
686 return 0;
689 /* fall back to mknod */
690 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
693 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
694 const char *dest)
696 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
697 struct ceph_mds_client *mdsc = fsc->mdsc;
698 struct ceph_mds_request *req;
699 int err;
701 if (ceph_snap(dir) != CEPH_NOSNAP)
702 return -EROFS;
704 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
705 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
706 if (IS_ERR(req)) {
707 d_drop(dentry);
708 return PTR_ERR(req);
710 req->r_dentry = dget(dentry);
711 req->r_num_caps = 2;
712 req->r_path2 = kstrdup(dest, GFP_NOFS);
713 req->r_locked_dir = dir;
714 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
715 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
716 err = ceph_mdsc_do_request(mdsc, dir, req);
717 if (!err && !req->r_reply_info.head->is_dentry)
718 err = ceph_handle_notrace_create(dir, dentry);
719 ceph_mdsc_put_request(req);
720 if (err)
721 d_drop(dentry);
722 return err;
725 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
727 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
728 struct ceph_mds_client *mdsc = fsc->mdsc;
729 struct ceph_mds_request *req;
730 int err = -EROFS;
731 int op;
733 if (ceph_snap(dir) == CEPH_SNAPDIR) {
734 /* mkdir .snap/foo is a MKSNAP */
735 op = CEPH_MDS_OP_MKSNAP;
736 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
737 dentry->d_name.len, dentry->d_name.name, dentry);
738 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
739 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
740 op = CEPH_MDS_OP_MKDIR;
741 } else {
742 goto out;
744 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
745 if (IS_ERR(req)) {
746 err = PTR_ERR(req);
747 goto out;
750 req->r_dentry = dget(dentry);
751 req->r_num_caps = 2;
752 req->r_locked_dir = dir;
753 req->r_args.mkdir.mode = cpu_to_le32(mode);
754 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
755 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
756 err = ceph_mdsc_do_request(mdsc, dir, req);
757 if (!err && !req->r_reply_info.head->is_dentry)
758 err = ceph_handle_notrace_create(dir, dentry);
759 ceph_mdsc_put_request(req);
760 out:
761 if (err < 0)
762 d_drop(dentry);
763 return err;
766 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
767 struct dentry *dentry)
769 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
770 struct ceph_mds_client *mdsc = fsc->mdsc;
771 struct ceph_mds_request *req;
772 int err;
774 if (ceph_snap(dir) != CEPH_NOSNAP)
775 return -EROFS;
777 dout("link in dir %p old_dentry %p dentry %p\n", dir,
778 old_dentry, dentry);
779 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
780 if (IS_ERR(req)) {
781 d_drop(dentry);
782 return PTR_ERR(req);
784 req->r_dentry = dget(dentry);
785 req->r_num_caps = 2;
786 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
787 req->r_locked_dir = dir;
788 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
789 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
790 err = ceph_mdsc_do_request(mdsc, dir, req);
791 if (err) {
792 d_drop(dentry);
793 } else if (!req->r_reply_info.head->is_dentry) {
794 ihold(old_dentry->d_inode);
795 d_instantiate(dentry, old_dentry->d_inode);
797 ceph_mdsc_put_request(req);
798 return err;
802 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
803 * looks like the link count will hit 0, drop any other caps (other
804 * than PIN) we don't specifically want (due to the file still being
805 * open).
807 static int drop_caps_for_unlink(struct inode *inode)
809 struct ceph_inode_info *ci = ceph_inode(inode);
810 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
812 spin_lock(&inode->i_lock);
813 if (inode->i_nlink == 1) {
814 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
815 ci->i_ceph_flags |= CEPH_I_NODELAY;
817 spin_unlock(&inode->i_lock);
818 return drop;
822 * rmdir and unlink are differ only by the metadata op code
824 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
826 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
827 struct ceph_mds_client *mdsc = fsc->mdsc;
828 struct inode *inode = dentry->d_inode;
829 struct ceph_mds_request *req;
830 int err = -EROFS;
831 int op;
833 if (ceph_snap(dir) == CEPH_SNAPDIR) {
834 /* rmdir .snap/foo is RMSNAP */
835 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
836 dentry->d_name.name, dentry);
837 op = CEPH_MDS_OP_RMSNAP;
838 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
839 dout("unlink/rmdir dir %p dn %p inode %p\n",
840 dir, dentry, inode);
841 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
842 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
843 } else
844 goto out;
845 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
846 if (IS_ERR(req)) {
847 err = PTR_ERR(req);
848 goto out;
850 req->r_dentry = dget(dentry);
851 req->r_num_caps = 2;
852 req->r_locked_dir = dir;
853 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
854 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
855 req->r_inode_drop = drop_caps_for_unlink(inode);
856 err = ceph_mdsc_do_request(mdsc, dir, req);
857 if (!err && !req->r_reply_info.head->is_dentry)
858 d_delete(dentry);
859 ceph_mdsc_put_request(req);
860 out:
861 return err;
864 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
865 struct inode *new_dir, struct dentry *new_dentry)
867 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
868 struct ceph_mds_client *mdsc = fsc->mdsc;
869 struct ceph_mds_request *req;
870 int err;
872 if (ceph_snap(old_dir) != ceph_snap(new_dir))
873 return -EXDEV;
874 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
875 ceph_snap(new_dir) != CEPH_NOSNAP)
876 return -EROFS;
877 dout("rename dir %p dentry %p to dir %p dentry %p\n",
878 old_dir, old_dentry, new_dir, new_dentry);
879 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
880 if (IS_ERR(req))
881 return PTR_ERR(req);
882 req->r_dentry = dget(new_dentry);
883 req->r_num_caps = 2;
884 req->r_old_dentry = dget(old_dentry);
885 req->r_locked_dir = new_dir;
886 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
887 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
888 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
889 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
890 /* release LINK_RDCACHE on source inode (mds will lock it) */
891 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
892 if (new_dentry->d_inode)
893 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
894 err = ceph_mdsc_do_request(mdsc, old_dir, req);
895 if (!err && !req->r_reply_info.head->is_dentry) {
897 * Normally d_move() is done by fill_trace (called by
898 * do_request, above). If there is no trace, we need
899 * to do it here.
902 /* d_move screws up d_subdirs order */
903 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
905 d_move(old_dentry, new_dentry);
907 /* ensure target dentry is invalidated, despite
908 rehashing bug in vfs_rename_dir */
909 ceph_invalidate_dentry_lease(new_dentry);
911 ceph_mdsc_put_request(req);
912 return err;
916 * Ensure a dentry lease will no longer revalidate.
918 void ceph_invalidate_dentry_lease(struct dentry *dentry)
920 spin_lock(&dentry->d_lock);
921 dentry->d_time = jiffies;
922 ceph_dentry(dentry)->lease_shared_gen = 0;
923 spin_unlock(&dentry->d_lock);
927 * Check if dentry lease is valid. If not, delete the lease. Try to
928 * renew if the least is more than half up.
930 static int dentry_lease_is_valid(struct dentry *dentry)
932 struct ceph_dentry_info *di;
933 struct ceph_mds_session *s;
934 int valid = 0;
935 u32 gen;
936 unsigned long ttl;
937 struct ceph_mds_session *session = NULL;
938 struct inode *dir = NULL;
939 u32 seq = 0;
941 spin_lock(&dentry->d_lock);
942 di = ceph_dentry(dentry);
943 if (di && di->lease_session) {
944 s = di->lease_session;
945 spin_lock(&s->s_cap_lock);
946 gen = s->s_cap_gen;
947 ttl = s->s_cap_ttl;
948 spin_unlock(&s->s_cap_lock);
950 if (di->lease_gen == gen &&
951 time_before(jiffies, dentry->d_time) &&
952 time_before(jiffies, ttl)) {
953 valid = 1;
954 if (di->lease_renew_after &&
955 time_after(jiffies, di->lease_renew_after)) {
956 /* we should renew */
957 dir = dentry->d_parent->d_inode;
958 session = ceph_get_mds_session(s);
959 seq = di->lease_seq;
960 di->lease_renew_after = 0;
961 di->lease_renew_from = jiffies;
965 spin_unlock(&dentry->d_lock);
967 if (session) {
968 ceph_mdsc_lease_send_msg(session, dir, dentry,
969 CEPH_MDS_LEASE_RENEW, seq);
970 ceph_put_mds_session(session);
972 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
973 return valid;
977 * Check if directory-wide content lease/cap is valid.
979 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
981 struct ceph_inode_info *ci = ceph_inode(dir);
982 struct ceph_dentry_info *di = ceph_dentry(dentry);
983 int valid = 0;
985 spin_lock(&dir->i_lock);
986 if (ci->i_shared_gen == di->lease_shared_gen)
987 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
988 spin_unlock(&dir->i_lock);
989 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
990 dir, (unsigned)ci->i_shared_gen, dentry,
991 (unsigned)di->lease_shared_gen, valid);
992 return valid;
996 * Check if cached dentry can be trusted.
998 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
1000 struct inode *dir;
1002 if (nd && nd->flags & LOOKUP_RCU)
1003 return -ECHILD;
1005 dir = dentry->d_parent->d_inode;
1007 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1008 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1009 ceph_dentry(dentry)->offset);
1011 /* always trust cached snapped dentries, snapdir dentry */
1012 if (ceph_snap(dir) != CEPH_NOSNAP) {
1013 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1014 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1015 goto out_touch;
1017 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1018 goto out_touch;
1020 if (dentry_lease_is_valid(dentry) ||
1021 dir_lease_is_valid(dir, dentry))
1022 goto out_touch;
1024 dout("d_revalidate %p invalid\n", dentry);
1025 d_drop(dentry);
1026 return 0;
1027 out_touch:
1028 ceph_dentry_lru_touch(dentry);
1029 return 1;
1033 * Release our ceph_dentry_info.
1035 static void ceph_d_release(struct dentry *dentry)
1037 struct ceph_dentry_info *di = ceph_dentry(dentry);
1039 dout("d_release %p\n", dentry);
1040 if (di) {
1041 ceph_dentry_lru_del(dentry);
1042 if (di->lease_session)
1043 ceph_put_mds_session(di->lease_session);
1044 kmem_cache_free(ceph_dentry_cachep, di);
1045 dentry->d_fsdata = NULL;
1049 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1050 struct nameidata *nd)
1053 * Eventually, we'll want to revalidate snapped metadata
1054 * too... probably...
1056 return 1;
1062 * read() on a dir. This weird interface hack only works if mounted
1063 * with '-o dirstat'.
1065 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1066 loff_t *ppos)
1068 struct ceph_file_info *cf = file->private_data;
1069 struct inode *inode = file->f_dentry->d_inode;
1070 struct ceph_inode_info *ci = ceph_inode(inode);
1071 int left;
1072 const int bufsize = 1024;
1074 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1075 return -EISDIR;
1077 if (!cf->dir_info) {
1078 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1079 if (!cf->dir_info)
1080 return -ENOMEM;
1081 cf->dir_info_len =
1082 snprintf(cf->dir_info, bufsize,
1083 "entries: %20lld\n"
1084 " files: %20lld\n"
1085 " subdirs: %20lld\n"
1086 "rentries: %20lld\n"
1087 " rfiles: %20lld\n"
1088 " rsubdirs: %20lld\n"
1089 "rbytes: %20lld\n"
1090 "rctime: %10ld.%09ld\n",
1091 ci->i_files + ci->i_subdirs,
1092 ci->i_files,
1093 ci->i_subdirs,
1094 ci->i_rfiles + ci->i_rsubdirs,
1095 ci->i_rfiles,
1096 ci->i_rsubdirs,
1097 ci->i_rbytes,
1098 (long)ci->i_rctime.tv_sec,
1099 (long)ci->i_rctime.tv_nsec);
1102 if (*ppos >= cf->dir_info_len)
1103 return 0;
1104 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1105 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1106 if (left == size)
1107 return -EFAULT;
1108 *ppos += (size - left);
1109 return size - left;
1113 * an fsync() on a dir will wait for any uncommitted directory
1114 * operations to commit.
1116 static int ceph_dir_fsync(struct file *file, int datasync)
1118 struct inode *inode = file->f_path.dentry->d_inode;
1119 struct ceph_inode_info *ci = ceph_inode(inode);
1120 struct list_head *head = &ci->i_unsafe_dirops;
1121 struct ceph_mds_request *req;
1122 u64 last_tid;
1123 int ret = 0;
1125 dout("dir_fsync %p\n", inode);
1126 spin_lock(&ci->i_unsafe_lock);
1127 if (list_empty(head))
1128 goto out;
1130 req = list_entry(head->prev,
1131 struct ceph_mds_request, r_unsafe_dir_item);
1132 last_tid = req->r_tid;
1134 do {
1135 ceph_mdsc_get_request(req);
1136 spin_unlock(&ci->i_unsafe_lock);
1137 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1138 inode, req->r_tid, last_tid);
1139 if (req->r_timeout) {
1140 ret = wait_for_completion_timeout(
1141 &req->r_safe_completion, req->r_timeout);
1142 if (ret > 0)
1143 ret = 0;
1144 else if (ret == 0)
1145 ret = -EIO; /* timed out */
1146 } else {
1147 wait_for_completion(&req->r_safe_completion);
1149 spin_lock(&ci->i_unsafe_lock);
1150 ceph_mdsc_put_request(req);
1152 if (ret || list_empty(head))
1153 break;
1154 req = list_entry(head->next,
1155 struct ceph_mds_request, r_unsafe_dir_item);
1156 } while (req->r_tid < last_tid);
1157 out:
1158 spin_unlock(&ci->i_unsafe_lock);
1159 return ret;
1163 * We maintain a private dentry LRU.
1165 * FIXME: this needs to be changed to a per-mds lru to be useful.
1167 void ceph_dentry_lru_add(struct dentry *dn)
1169 struct ceph_dentry_info *di = ceph_dentry(dn);
1170 struct ceph_mds_client *mdsc;
1172 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1173 dn->d_name.len, dn->d_name.name);
1174 if (di) {
1175 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1176 spin_lock(&mdsc->dentry_lru_lock);
1177 list_add_tail(&di->lru, &mdsc->dentry_lru);
1178 mdsc->num_dentry++;
1179 spin_unlock(&mdsc->dentry_lru_lock);
1183 void ceph_dentry_lru_touch(struct dentry *dn)
1185 struct ceph_dentry_info *di = ceph_dentry(dn);
1186 struct ceph_mds_client *mdsc;
1188 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1189 dn->d_name.len, dn->d_name.name, di->offset);
1190 if (di) {
1191 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1192 spin_lock(&mdsc->dentry_lru_lock);
1193 list_move_tail(&di->lru, &mdsc->dentry_lru);
1194 spin_unlock(&mdsc->dentry_lru_lock);
1198 void ceph_dentry_lru_del(struct dentry *dn)
1200 struct ceph_dentry_info *di = ceph_dentry(dn);
1201 struct ceph_mds_client *mdsc;
1203 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1204 dn->d_name.len, dn->d_name.name);
1205 if (di) {
1206 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1207 spin_lock(&mdsc->dentry_lru_lock);
1208 list_del_init(&di->lru);
1209 mdsc->num_dentry--;
1210 spin_unlock(&mdsc->dentry_lru_lock);
1215 * Return name hash for a given dentry. This is dependent on
1216 * the parent directory's hash function.
1218 unsigned ceph_dentry_hash(struct dentry *dn)
1220 struct inode *dir = dn->d_parent->d_inode;
1221 struct ceph_inode_info *dci = ceph_inode(dir);
1223 switch (dci->i_dir_layout.dl_dir_hash) {
1224 case 0: /* for backward compat */
1225 case CEPH_STR_HASH_LINUX:
1226 return dn->d_name.hash;
1228 default:
1229 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1230 dn->d_name.name, dn->d_name.len);
1234 const struct file_operations ceph_dir_fops = {
1235 .read = ceph_read_dir,
1236 .readdir = ceph_readdir,
1237 .llseek = ceph_dir_llseek,
1238 .open = ceph_open,
1239 .release = ceph_release,
1240 .unlocked_ioctl = ceph_ioctl,
1241 .fsync = ceph_dir_fsync,
1244 const struct inode_operations ceph_dir_iops = {
1245 .lookup = ceph_lookup,
1246 .permission = ceph_permission,
1247 .getattr = ceph_getattr,
1248 .setattr = ceph_setattr,
1249 .setxattr = ceph_setxattr,
1250 .getxattr = ceph_getxattr,
1251 .listxattr = ceph_listxattr,
1252 .removexattr = ceph_removexattr,
1253 .mknod = ceph_mknod,
1254 .symlink = ceph_symlink,
1255 .mkdir = ceph_mkdir,
1256 .link = ceph_link,
1257 .unlink = ceph_unlink,
1258 .rmdir = ceph_unlink,
1259 .rename = ceph_rename,
1260 .create = ceph_create,
1263 const struct dentry_operations ceph_dentry_ops = {
1264 .d_revalidate = ceph_d_revalidate,
1265 .d_release = ceph_d_release,
1268 const struct dentry_operations ceph_snapdir_dentry_ops = {
1269 .d_revalidate = ceph_snapdir_d_revalidate,
1270 .d_release = ceph_d_release,
1273 const struct dentry_operations ceph_snap_dentry_ops = {
1274 .d_release = ceph_d_release,