blob: 3e8094be4604a481999e2921fb0942441c9086a7 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil2817b002009-10-06 11:31:08 -07002
3#include <linux/spinlock.h>
4#include <linux/fs_struct.h>
5#include <linux/namei.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09006#include <linux/slab.h>
Sage Weil2817b002009-10-06 11:31:08 -07007#include <linux/sched.h>
8
9#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070010#include "mds_client.h"
Sage Weil2817b002009-10-06 11:31:08 -070011
12/*
13 * Directory operations: readdir, lookup, create, link, unlink,
14 * rename, etc.
15 */
16
17/*
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.
22 *
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.
27 */
28
29const struct inode_operations ceph_dir_iops;
30const struct file_operations ceph_dir_fops;
Sage Weil52dfb8a2010-08-03 10:25:30 -070031const struct dentry_operations ceph_dentry_ops;
Sage Weil2817b002009-10-06 11:31:08 -070032
33/*
34 * Initialize ceph dentry state.
35 */
36int ceph_init_dentry(struct dentry *dentry)
37{
38 struct ceph_dentry_info *di;
39
40 if (dentry->d_fsdata)
41 return 0;
42
Sage Weil36e21682010-08-24 16:23:48 -070043 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
Sage Weil2817b002009-10-06 11:31:08 -070044 if (!di)
45 return -ENOMEM; /* oh well */
46
47 spin_lock(&dentry->d_lock);
Sage Weil8c6efb52010-04-23 11:36:54 -070048 if (dentry->d_fsdata) {
49 /* lost a race */
50 kmem_cache_free(ceph_dentry_cachep, di);
Sage Weil2817b002009-10-06 11:31:08 -070051 goto out_unlock;
Sage Weil8c6efb52010-04-23 11:36:54 -070052 }
Sage Weil48d0cbd2011-07-26 11:30:15 -070053
54 if (dentry->d_parent == NULL || /* nfs fh_to_dentry */
55 ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
56 d_set_d_op(dentry, &ceph_dentry_ops);
57 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
58 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
59 else
60 d_set_d_op(dentry, &ceph_snap_dentry_ops);
61
Sage Weil2817b002009-10-06 11:31:08 -070062 di->dentry = dentry;
63 di->lease_session = NULL;
Sage Weil2817b002009-10-06 11:31:08 -070064 dentry->d_time = jiffies;
Sage Weil48d0cbd2011-07-26 11:30:15 -070065 /* avoid reordering d_fsdata setup so that the check above is safe */
66 smp_mb();
67 dentry->d_fsdata = di;
Sage Weil2817b002009-10-06 11:31:08 -070068 ceph_dentry_lru_add(dentry);
69out_unlock:
70 spin_unlock(&dentry->d_lock);
71 return 0;
72}
73
Sage Weil5f21c962011-07-26 11:30:29 -070074struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
75{
76 struct inode *inode = NULL;
77
78 if (!dentry)
79 return NULL;
80
81 spin_lock(&dentry->d_lock);
82 if (dentry->d_parent) {
83 inode = dentry->d_parent->d_inode;
84 ihold(inode);
85 }
86 spin_unlock(&dentry->d_lock);
87 return inode;
88}
Sage Weil2817b002009-10-06 11:31:08 -070089
90
91/*
92 * for readdir, we encode the directory frag and offset within that
93 * frag into f_pos.
94 */
95static unsigned fpos_frag(loff_t p)
96{
97 return p >> 32;
98}
99static unsigned fpos_off(loff_t p)
100{
101 return p & 0xffffffff;
102}
103
104/*
105 * When possible, we try to satisfy a readdir by peeking at the
106 * dcache. We make this work by carefully ordering dentries on
107 * d_u.d_child when we initially get results back from the MDS, and
108 * falling back to a "normal" sync readdir if any dentries in the dir
109 * are dropped.
110 *
Sage Weilc6ffe102011-11-03 09:23:36 -0700111 * D_COMPLETE tells indicates we have all dentries in the dir. It is
Sage Weil2817b002009-10-06 11:31:08 -0700112 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
113 * the MDS if/when the directory is modified).
114 */
115static int __dcache_readdir(struct file *filp,
116 void *dirent, filldir_t filldir)
117{
Sage Weil2817b002009-10-06 11:31:08 -0700118 struct ceph_file_info *fi = filp->private_data;
119 struct dentry *parent = filp->f_dentry;
120 struct inode *dir = parent->d_inode;
121 struct list_head *p;
122 struct dentry *dentry, *last;
123 struct ceph_dentry_info *di;
124 int err = 0;
125
126 /* claim ref on last dentry we returned */
127 last = fi->dentry;
128 fi->dentry = NULL;
129
130 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
131 last);
132
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100133 spin_lock(&parent->d_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700134
135 /* start at beginning? */
Sage Weil884ea892010-11-22 22:58:06 -0800136 if (filp->f_pos == 2 || last == NULL ||
137 filp->f_pos < ceph_dentry(last)->offset) {
Sage Weil2817b002009-10-06 11:31:08 -0700138 if (list_empty(&parent->d_subdirs))
139 goto out_unlock;
140 p = parent->d_subdirs.prev;
141 dout(" initial p %p/%p\n", p->prev, p->next);
142 } else {
143 p = last->d_u.d_child.prev;
144 }
145
146more:
147 dentry = list_entry(p, struct dentry, d_u.d_child);
148 di = ceph_dentry(dentry);
149 while (1) {
Sage Weil1cd39352010-05-03 22:08:02 -0700150 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
151 d_unhashed(dentry) ? "!hashed" : "hashed",
Sage Weil2817b002009-10-06 11:31:08 -0700152 parent->d_subdirs.prev, parent->d_subdirs.next);
153 if (p == &parent->d_subdirs) {
Sage Weil9cfa1092011-07-26 11:26:18 -0700154 fi->flags |= CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700155 goto out_unlock;
156 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100157 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Sage Weil2817b002009-10-06 11:31:08 -0700158 if (!d_unhashed(dentry) && dentry->d_inode &&
Sage Weil09b8a7d2009-11-11 15:21:27 -0800159 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
Sage Weil1d1de9162009-12-02 11:54:25 -0800160 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
Sage Weil2817b002009-10-06 11:31:08 -0700161 filp->f_pos <= di->offset)
162 break;
163 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
164 dentry->d_name.len, dentry->d_name.name, di->offset,
165 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
166 !dentry->d_inode ? " null" : "");
Nick Pigginda502952011-01-07 17:49:33 +1100167 spin_unlock(&dentry->d_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700168 p = p->prev;
169 dentry = list_entry(p, struct dentry, d_u.d_child);
170 di = ceph_dentry(dentry);
171 }
172
Nick Pigginda502952011-01-07 17:49:33 +1100173 dget_dlock(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100174 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100175 spin_unlock(&parent->d_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700176
177 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
178 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
179 filp->f_pos = di->offset;
180 err = filldir(dirent, dentry->d_name.name,
181 dentry->d_name.len, di->offset,
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800182 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
Sage Weil2817b002009-10-06 11:31:08 -0700183 dentry->d_inode->i_mode >> 12);
184
185 if (last) {
186 if (err < 0) {
187 /* remember our position */
188 fi->dentry = last;
189 fi->next_offset = di->offset;
190 } else {
191 dput(last);
192 }
Sage Weil2817b002009-10-06 11:31:08 -0700193 }
Sage Weilf5b06622010-04-12 14:24:28 -0700194 last = dentry;
195
Sage Weil2817b002009-10-06 11:31:08 -0700196 if (err < 0)
Sage Weilefa4c122010-10-18 14:04:31 -0700197 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700198
Sage Weil2817b002009-10-06 11:31:08 -0700199 filp->f_pos++;
200
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100201 /* make sure a dentry wasn't dropped while we didn't have parent lock */
Sage Weilc6ffe102011-11-03 09:23:36 -0700202 if (!ceph_dir_test_complete(dir)) {
203 dout(" lost D_COMPLETE on %p; falling back to mds\n", dir);
Sage Weilefa4c122010-10-18 14:04:31 -0700204 err = -EAGAIN;
205 goto out;
206 }
207
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100208 spin_lock(&parent->d_lock);
Sage Weilefa4c122010-10-18 14:04:31 -0700209 p = p->prev; /* advance to next dentry */
210 goto more;
Sage Weil2817b002009-10-06 11:31:08 -0700211
212out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100213 spin_unlock(&parent->d_lock);
Sage Weilefa4c122010-10-18 14:04:31 -0700214out:
215 if (last)
Sage Weil2817b002009-10-06 11:31:08 -0700216 dput(last);
Sage Weil2817b002009-10-06 11:31:08 -0700217 return err;
218}
219
220/*
221 * make note of the last dentry we read, so we can
222 * continue at the same lexicographical point,
223 * regardless of what dir changes take place on the
224 * server.
225 */
226static int note_last_dentry(struct ceph_file_info *fi, const char *name,
227 int len)
228{
229 kfree(fi->last_name);
230 fi->last_name = kmalloc(len+1, GFP_NOFS);
231 if (!fi->last_name)
232 return -ENOMEM;
233 memcpy(fi->last_name, name, len);
234 fi->last_name[len] = 0;
235 dout("note_last_dentry '%s'\n", fi->last_name);
236 return 0;
237}
238
239static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
240{
241 struct ceph_file_info *fi = filp->private_data;
242 struct inode *inode = filp->f_dentry->d_inode;
243 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700244 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
245 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700246 unsigned frag = fpos_frag(filp->f_pos);
247 int off = fpos_off(filp->f_pos);
248 int err;
249 u32 ftype;
250 struct ceph_mds_reply_info_parsed *rinfo;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700251 const int max_entries = fsc->mount_options->max_readdir;
252 const int max_bytes = fsc->mount_options->max_readdir_bytes;
Sage Weil2817b002009-10-06 11:31:08 -0700253
254 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
Sage Weil9cfa1092011-07-26 11:26:18 -0700255 if (fi->flags & CEPH_F_ATEND)
Sage Weil2817b002009-10-06 11:31:08 -0700256 return 0;
257
258 /* always start with . and .. */
259 if (filp->f_pos == 0) {
260 /* note dir version at start of readdir so we can tell
261 * if any dentries get dropped */
262 fi->dir_release_count = ci->i_release_count;
263
264 dout("readdir off 0 -> '.'\n");
265 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800266 ceph_translate_ino(inode->i_sb, inode->i_ino),
267 inode->i_mode >> 12) < 0)
Sage Weil2817b002009-10-06 11:31:08 -0700268 return 0;
269 filp->f_pos = 1;
270 off = 1;
271 }
272 if (filp->f_pos == 1) {
Al Virob85fd6b2011-07-17 11:19:44 -0400273 ino_t ino = parent_ino(filp->f_dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700274 dout("readdir off 1 -> '..'\n");
275 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800276 ceph_translate_ino(inode->i_sb, ino),
Sage Weil2817b002009-10-06 11:31:08 -0700277 inode->i_mode >> 12) < 0)
278 return 0;
279 filp->f_pos = 2;
280 off = 2;
281 }
282
283 /* can we use the dcache? */
Sage Weilbe655592011-11-30 09:47:09 -0800284 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700285 if ((filp->f_pos == 2 || fi->dentry) &&
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700286 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
Sage Weila0dff782010-07-22 13:47:21 -0700287 ceph_snap(inode) != CEPH_SNAPDIR &&
Sage Weilc6ffe102011-11-03 09:23:36 -0700288 ceph_dir_test_complete(inode) &&
Sage Weil2817b002009-10-06 11:31:08 -0700289 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
Sage Weilbe655592011-11-30 09:47:09 -0800290 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700291 err = __dcache_readdir(filp, dirent, filldir);
Sage Weilefa4c122010-10-18 14:04:31 -0700292 if (err != -EAGAIN)
Sage Weil2817b002009-10-06 11:31:08 -0700293 return err;
Sage Weilefa4c122010-10-18 14:04:31 -0700294 } else {
Sage Weilbe655592011-11-30 09:47:09 -0800295 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700296 }
Sage Weil2817b002009-10-06 11:31:08 -0700297 if (fi->dentry) {
298 err = note_last_dentry(fi, fi->dentry->d_name.name,
299 fi->dentry->d_name.len);
300 if (err)
301 return err;
302 dput(fi->dentry);
303 fi->dentry = NULL;
304 }
305
306 /* proceed with a normal readdir */
307
308more:
309 /* do we have the correct frag content buffered? */
310 if (fi->frag != frag || fi->last_readdir == NULL) {
311 struct ceph_mds_request *req;
312 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
313 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
314
315 /* discard old result, if any */
Sage Weil393f6622010-03-10 12:03:32 -0800316 if (fi->last_readdir) {
Sage Weil2817b002009-10-06 11:31:08 -0700317 ceph_mdsc_put_request(fi->last_readdir);
Sage Weil393f6622010-03-10 12:03:32 -0800318 fi->last_readdir = NULL;
319 }
Sage Weil2817b002009-10-06 11:31:08 -0700320
321 /* requery frag tree, as the frag topology may have changed */
322 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
323
324 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
325 ceph_vinop(inode), frag, fi->last_name);
326 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
327 if (IS_ERR(req))
328 return PTR_ERR(req);
Sage Weil70b666c2011-05-27 09:24:26 -0700329 req->r_inode = inode;
330 ihold(inode);
Sage Weil2817b002009-10-06 11:31:08 -0700331 req->r_dentry = dget(filp->f_dentry);
332 /* hints to request -> mds selection code */
333 req->r_direct_mode = USE_AUTH_MDS;
334 req->r_direct_hash = ceph_frag_value(frag);
335 req->r_direct_is_hash = true;
336 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
337 req->r_readdir_offset = fi->next_offset;
338 req->r_args.readdir.frag = cpu_to_le32(frag);
339 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
Sage Weil23804d92010-05-14 13:06:30 -0700340 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
Yehuda Sadehe1e4dd02010-04-13 11:45:56 -0700341 req->r_num_caps = max_entries + 1;
Sage Weil2817b002009-10-06 11:31:08 -0700342 err = ceph_mdsc_do_request(mdsc, NULL, req);
343 if (err < 0) {
344 ceph_mdsc_put_request(req);
345 return err;
346 }
347 dout("readdir got and parsed readdir result=%d"
348 " on frag %x, end=%d, complete=%d\n", err, frag,
349 (int)req->r_reply_info.dir_end,
350 (int)req->r_reply_info.dir_complete);
351
352 if (!req->r_did_prepopulate) {
353 dout("readdir !did_prepopulate");
Sage Weilc6ffe102011-11-03 09:23:36 -0700354 fi->dir_release_count--; /* preclude D_COMPLETE */
Sage Weil2817b002009-10-06 11:31:08 -0700355 }
356
357 /* note next offset and last dentry name */
358 fi->offset = fi->next_offset;
359 fi->last_readdir = req;
360
361 if (req->r_reply_info.dir_end) {
362 kfree(fi->last_name);
363 fi->last_name = NULL;
Sage Weil7b88dad2010-11-11 16:48:59 -0800364 if (ceph_frag_is_rightmost(frag))
365 fi->next_offset = 2;
366 else
367 fi->next_offset = 0;
Sage Weil2817b002009-10-06 11:31:08 -0700368 } else {
369 rinfo = &req->r_reply_info;
370 err = note_last_dentry(fi,
371 rinfo->dir_dname[rinfo->dir_nr-1],
372 rinfo->dir_dname_len[rinfo->dir_nr-1]);
373 if (err)
374 return err;
375 fi->next_offset += rinfo->dir_nr;
376 }
377 }
378
379 rinfo = &fi->last_readdir->r_reply_info;
380 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
381 rinfo->dir_nr, off, fi->offset);
Sage Weilda398222011-05-12 15:28:11 -0700382 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
Sage Weil2817b002009-10-06 11:31:08 -0700383 u64 pos = ceph_make_fpos(frag, off);
384 struct ceph_mds_reply_inode *in =
385 rinfo->dir_in[off - fi->offset].in;
Sage Weil3105c192010-11-18 09:15:07 -0800386 struct ceph_vino vino;
387 ino_t ino;
388
Sage Weil2817b002009-10-06 11:31:08 -0700389 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
390 off, off - fi->offset, rinfo->dir_nr, pos,
391 rinfo->dir_dname_len[off - fi->offset],
392 rinfo->dir_dname[off - fi->offset], in);
393 BUG_ON(!in);
394 ftype = le32_to_cpu(in->mode) >> 12;
Sage Weil3105c192010-11-18 09:15:07 -0800395 vino.ino = le64_to_cpu(in->ino);
396 vino.snap = le64_to_cpu(in->snapid);
397 ino = ceph_vino_to_ino(vino);
Sage Weil2817b002009-10-06 11:31:08 -0700398 if (filldir(dirent,
399 rinfo->dir_dname[off - fi->offset],
400 rinfo->dir_dname_len[off - fi->offset],
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800401 pos,
402 ceph_translate_ino(inode->i_sb, ino), ftype) < 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700403 dout("filldir stopping us...\n");
404 return 0;
405 }
406 off++;
407 filp->f_pos = pos + 1;
408 }
409
410 if (fi->last_name) {
411 ceph_mdsc_put_request(fi->last_readdir);
412 fi->last_readdir = NULL;
413 goto more;
414 }
415
416 /* more frags? */
417 if (!ceph_frag_is_rightmost(frag)) {
418 frag = ceph_frag_next(frag);
419 off = 0;
420 filp->f_pos = ceph_make_fpos(frag, off);
421 dout("readdir next frag is %x\n", frag);
422 goto more;
423 }
Sage Weil9cfa1092011-07-26 11:26:18 -0700424 fi->flags |= CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700425
426 /*
427 * if dir_release_count still matches the dir, no dentries
428 * were released during the whole readdir, and we should have
429 * the complete dir contents in our cache.
430 */
Sage Weilbe655592011-11-30 09:47:09 -0800431 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700432 if (ci->i_release_count == fi->dir_release_count) {
Sage Weilc6ffe102011-11-03 09:23:36 -0700433 ceph_dir_set_complete(inode);
Sage Weil2817b002009-10-06 11:31:08 -0700434 ci->i_max_offset = filp->f_pos;
435 }
Sage Weilbe655592011-11-30 09:47:09 -0800436 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700437
438 dout("readdir %p filp %p done.\n", inode, filp);
439 return 0;
440}
441
442static void reset_readdir(struct ceph_file_info *fi)
443{
444 if (fi->last_readdir) {
445 ceph_mdsc_put_request(fi->last_readdir);
446 fi->last_readdir = NULL;
447 }
448 kfree(fi->last_name);
Sage Weila1629c32010-11-11 15:24:06 -0800449 fi->last_name = NULL;
Sage Weil2817b002009-10-06 11:31:08 -0700450 fi->next_offset = 2; /* compensate for . and .. */
451 if (fi->dentry) {
452 dput(fi->dentry);
453 fi->dentry = NULL;
454 }
Sage Weil9cfa1092011-07-26 11:26:18 -0700455 fi->flags &= ~CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700456}
457
458static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
459{
460 struct ceph_file_info *fi = file->private_data;
461 struct inode *inode = file->f_mapping->host;
462 loff_t old_offset = offset;
463 loff_t retval;
464
465 mutex_lock(&inode->i_mutex);
Josef Bacik06222e42011-07-18 13:21:38 -0400466 retval = -EINVAL;
Sage Weil2817b002009-10-06 11:31:08 -0700467 switch (origin) {
468 case SEEK_END:
469 offset += inode->i_size + 2; /* FIXME */
470 break;
471 case SEEK_CUR:
472 offset += file->f_pos;
Josef Bacik06222e42011-07-18 13:21:38 -0400473 case SEEK_SET:
474 break;
475 default:
476 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700477 }
Josef Bacik06222e42011-07-18 13:21:38 -0400478
Sage Weil2817b002009-10-06 11:31:08 -0700479 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
480 if (offset != file->f_pos) {
481 file->f_pos = offset;
482 file->f_version = 0;
Sage Weil9cfa1092011-07-26 11:26:18 -0700483 fi->flags &= ~CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700484 }
485 retval = offset;
486
487 /*
488 * discard buffered readdir content on seekdir(0), or
489 * seek to new frag, or seek prior to current chunk.
490 */
491 if (offset == 0 ||
492 fpos_frag(offset) != fpos_frag(old_offset) ||
493 fpos_off(offset) < fi->offset) {
494 dout("dir_llseek dropping %p content\n", file);
495 reset_readdir(fi);
496 }
497
498 /* bump dir_release_count if we did a forward seek */
499 if (offset > old_offset)
500 fi->dir_release_count--;
501 }
Josef Bacik06222e42011-07-18 13:21:38 -0400502out:
Sage Weil2817b002009-10-06 11:31:08 -0700503 mutex_unlock(&inode->i_mutex);
504 return retval;
505}
506
507/*
Sage Weil468640e2011-07-26 11:28:11 -0700508 * Handle lookups for the hidden .snap directory.
Sage Weil2817b002009-10-06 11:31:08 -0700509 */
Sage Weil468640e2011-07-26 11:28:11 -0700510int ceph_handle_snapdir(struct ceph_mds_request *req,
511 struct dentry *dentry, int err)
Sage Weil2817b002009-10-06 11:31:08 -0700512{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700513 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
Sage Weild79698d2011-07-26 11:31:26 -0700514 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
Sage Weil2817b002009-10-06 11:31:08 -0700515
516 /* .snap dir? */
517 if (err == -ENOENT &&
Sage Weil455cec02011-03-03 13:44:35 -0800518 ceph_snap(parent) == CEPH_NOSNAP &&
Sage Weil6b805182009-10-27 11:50:50 -0700519 strcmp(dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700520 fsc->mount_options->snapdir_name) == 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700521 struct inode *inode = ceph_get_snapdir(parent);
522 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
523 dentry, dentry->d_name.len, dentry->d_name.name, inode);
Sage Weil9358c6d2010-03-30 13:54:41 -0700524 BUG_ON(!d_unhashed(dentry));
Sage Weil2817b002009-10-06 11:31:08 -0700525 d_add(dentry, inode);
526 err = 0;
527 }
Sage Weil468640e2011-07-26 11:28:11 -0700528 return err;
529}
Sage Weil2817b002009-10-06 11:31:08 -0700530
Sage Weil468640e2011-07-26 11:28:11 -0700531/*
532 * Figure out final result of a lookup/open request.
533 *
534 * Mainly, make sure we return the final req->r_dentry (if it already
535 * existed) in place of the original VFS-provided dentry when they
536 * differ.
537 *
538 * Gracefully handle the case where the MDS replies with -ENOENT and
539 * no trace (which it may do, at its discretion, e.g., if it doesn't
540 * care to issue a lease on the negative dentry).
541 */
542struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
543 struct dentry *dentry, int err)
544{
Sage Weil2817b002009-10-06 11:31:08 -0700545 if (err == -ENOENT) {
546 /* no trace? */
547 err = 0;
548 if (!req->r_reply_info.head->is_dentry) {
549 dout("ENOENT and no trace, dentry %p inode %p\n",
550 dentry, dentry->d_inode);
551 if (dentry->d_inode) {
552 d_drop(dentry);
553 err = -ENOENT;
554 } else {
555 d_add(dentry, NULL);
556 }
557 }
558 }
559 if (err)
560 dentry = ERR_PTR(err);
561 else if (dentry != req->r_dentry)
562 dentry = dget(req->r_dentry); /* we got spliced */
563 else
564 dentry = NULL;
565 return dentry;
566}
567
Sage Weil1d1de9162009-12-02 11:54:25 -0800568static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
569{
570 return ceph_ino(inode) == CEPH_INO_ROOT &&
571 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
572}
573
Sage Weil2817b002009-10-06 11:31:08 -0700574/*
575 * Look up a single dir entry. If there is a lookup intent, inform
576 * the MDS so that it gets our 'caps wanted' value in a single op.
577 */
578static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
579 struct nameidata *nd)
580{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700581 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
582 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700583 struct ceph_mds_request *req;
584 int op;
585 int err;
586
587 dout("lookup %p dentry %p '%.*s'\n",
588 dir, dentry, dentry->d_name.len, dentry->d_name.name);
589
590 if (dentry->d_name.len > NAME_MAX)
591 return ERR_PTR(-ENAMETOOLONG);
592
593 err = ceph_init_dentry(dentry);
594 if (err < 0)
595 return ERR_PTR(err);
596
597 /* open (but not create!) intent? */
598 if (nd &&
599 (nd->flags & LOOKUP_OPEN) &&
Sage Weil2817b002009-10-06 11:31:08 -0700600 !(nd->intent.open.flags & O_CREAT)) {
601 int mode = nd->intent.open.create_mode & ~current->fs->umask;
602 return ceph_lookup_open(dir, dentry, nd, mode, 1);
603 }
604
605 /* can we conclude ENOENT locally? */
606 if (dentry->d_inode == NULL) {
607 struct ceph_inode_info *ci = ceph_inode(dir);
608 struct ceph_dentry_info *di = ceph_dentry(dentry);
609
Sage Weilbe655592011-11-30 09:47:09 -0800610 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700611 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
612 if (strncmp(dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700613 fsc->mount_options->snapdir_name,
Sage Weil2817b002009-10-06 11:31:08 -0700614 dentry->d_name.len) &&
Sage Weil1d1de9162009-12-02 11:54:25 -0800615 !is_root_ceph_dentry(dir, dentry) &&
Sage Weilc6ffe102011-11-03 09:23:36 -0700616 ceph_dir_test_complete(dir) &&
Sage Weil2817b002009-10-06 11:31:08 -0700617 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
Sage Weilbe655592011-11-30 09:47:09 -0800618 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700619 dout(" dir %p complete, -ENOENT\n", dir);
620 d_add(dentry, NULL);
621 di->lease_shared_gen = ci->i_shared_gen;
622 return NULL;
623 }
Sage Weilbe655592011-11-30 09:47:09 -0800624 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700625 }
626
627 op = ceph_snap(dir) == CEPH_SNAPDIR ?
628 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
629 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
630 if (IS_ERR(req))
Julia Lawall7e34bc52010-05-22 12:01:14 +0200631 return ERR_CAST(req);
Sage Weil2817b002009-10-06 11:31:08 -0700632 req->r_dentry = dget(dentry);
633 req->r_num_caps = 2;
634 /* we only need inode linkage */
635 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
636 req->r_locked_dir = dir;
637 err = ceph_mdsc_do_request(mdsc, NULL, req);
Sage Weil468640e2011-07-26 11:28:11 -0700638 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil2817b002009-10-06 11:31:08 -0700639 dentry = ceph_finish_lookup(req, dentry, err);
640 ceph_mdsc_put_request(req); /* will dput(dentry) */
641 dout("lookup result=%p\n", dentry);
642 return dentry;
643}
644
645/*
646 * If we do a create but get no trace back from the MDS, follow up with
647 * a lookup (the VFS expects us to link up the provided dentry).
648 */
649int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
650{
651 struct dentry *result = ceph_lookup(dir, dentry, NULL);
652
653 if (result && !IS_ERR(result)) {
654 /*
655 * We created the item, then did a lookup, and found
656 * it was already linked to another inode we already
657 * had in our cache (and thus got spliced). Link our
658 * dentry to that inode, but don't hash it, just in
659 * case the VFS wants to dereference it.
660 */
661 BUG_ON(!result->d_inode);
662 d_instantiate(dentry, result->d_inode);
663 return 0;
664 }
665 return PTR_ERR(result);
666}
667
668static int ceph_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -0400669 umode_t mode, dev_t rdev)
Sage Weil2817b002009-10-06 11:31:08 -0700670{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700671 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
672 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700673 struct ceph_mds_request *req;
674 int err;
675
676 if (ceph_snap(dir) != CEPH_NOSNAP)
677 return -EROFS;
678
Al Viro1a67aaf2011-07-26 01:52:52 -0400679 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
Sage Weil2817b002009-10-06 11:31:08 -0700680 dir, dentry, mode, rdev);
681 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
682 if (IS_ERR(req)) {
683 d_drop(dentry);
684 return PTR_ERR(req);
685 }
686 req->r_dentry = dget(dentry);
687 req->r_num_caps = 2;
688 req->r_locked_dir = dir;
689 req->r_args.mknod.mode = cpu_to_le32(mode);
690 req->r_args.mknod.rdev = cpu_to_le32(rdev);
691 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
692 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
693 err = ceph_mdsc_do_request(mdsc, dir, req);
694 if (!err && !req->r_reply_info.head->is_dentry)
695 err = ceph_handle_notrace_create(dir, dentry);
696 ceph_mdsc_put_request(req);
697 if (err)
698 d_drop(dentry);
699 return err;
700}
701
Al Viro4acdaf22011-07-26 01:42:34 -0400702static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Sage Weil2817b002009-10-06 11:31:08 -0700703 struct nameidata *nd)
704{
705 dout("create in dir %p dentry %p name '%.*s'\n",
706 dir, dentry, dentry->d_name.len, dentry->d_name.name);
707
708 if (ceph_snap(dir) != CEPH_NOSNAP)
709 return -EROFS;
710
711 if (nd) {
712 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
713 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
714 /* hrm, what should i do here if we get aliased? */
715 if (IS_ERR(dentry))
716 return PTR_ERR(dentry);
717 return 0;
718 }
719
720 /* fall back to mknod */
721 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
722}
723
724static int ceph_symlink(struct inode *dir, struct dentry *dentry,
725 const char *dest)
726{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700727 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
728 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700729 struct ceph_mds_request *req;
730 int err;
731
732 if (ceph_snap(dir) != CEPH_NOSNAP)
733 return -EROFS;
734
735 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
736 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
737 if (IS_ERR(req)) {
738 d_drop(dentry);
739 return PTR_ERR(req);
740 }
741 req->r_dentry = dget(dentry);
742 req->r_num_caps = 2;
743 req->r_path2 = kstrdup(dest, GFP_NOFS);
744 req->r_locked_dir = dir;
745 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
746 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
747 err = ceph_mdsc_do_request(mdsc, dir, req);
748 if (!err && !req->r_reply_info.head->is_dentry)
749 err = ceph_handle_notrace_create(dir, dentry);
750 ceph_mdsc_put_request(req);
751 if (err)
752 d_drop(dentry);
753 return err;
754}
755
Al Viro18bb1db2011-07-26 01:41:39 -0400756static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Sage Weil2817b002009-10-06 11:31:08 -0700757{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700758 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
759 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700760 struct ceph_mds_request *req;
761 int err = -EROFS;
762 int op;
763
764 if (ceph_snap(dir) == CEPH_SNAPDIR) {
765 /* mkdir .snap/foo is a MKSNAP */
766 op = CEPH_MDS_OP_MKSNAP;
767 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
768 dentry->d_name.len, dentry->d_name.name, dentry);
769 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
Al Viro18bb1db2011-07-26 01:41:39 -0400770 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
Sage Weil2817b002009-10-06 11:31:08 -0700771 op = CEPH_MDS_OP_MKDIR;
772 } else {
773 goto out;
774 }
775 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
776 if (IS_ERR(req)) {
777 err = PTR_ERR(req);
778 goto out;
779 }
780
781 req->r_dentry = dget(dentry);
782 req->r_num_caps = 2;
783 req->r_locked_dir = dir;
784 req->r_args.mkdir.mode = cpu_to_le32(mode);
785 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
786 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
787 err = ceph_mdsc_do_request(mdsc, dir, req);
788 if (!err && !req->r_reply_info.head->is_dentry)
789 err = ceph_handle_notrace_create(dir, dentry);
790 ceph_mdsc_put_request(req);
791out:
792 if (err < 0)
793 d_drop(dentry);
794 return err;
795}
796
797static int ceph_link(struct dentry *old_dentry, struct inode *dir,
798 struct dentry *dentry)
799{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700800 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
801 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700802 struct ceph_mds_request *req;
803 int err;
804
805 if (ceph_snap(dir) != CEPH_NOSNAP)
806 return -EROFS;
807
808 dout("link in dir %p old_dentry %p dentry %p\n", dir,
809 old_dentry, dentry);
810 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
811 if (IS_ERR(req)) {
812 d_drop(dentry);
813 return PTR_ERR(req);
814 }
815 req->r_dentry = dget(dentry);
816 req->r_num_caps = 2;
817 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
Sage Weil41b02e12011-07-26 11:31:14 -0700818 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700819 req->r_locked_dir = dir;
820 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
821 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
822 err = ceph_mdsc_do_request(mdsc, dir, req);
Sage Weil70b666c2011-05-27 09:24:26 -0700823 if (err) {
Sage Weil2817b002009-10-06 11:31:08 -0700824 d_drop(dentry);
Sage Weil70b666c2011-05-27 09:24:26 -0700825 } else if (!req->r_reply_info.head->is_dentry) {
826 ihold(old_dentry->d_inode);
827 d_instantiate(dentry, old_dentry->d_inode);
828 }
Sage Weil2817b002009-10-06 11:31:08 -0700829 ceph_mdsc_put_request(req);
830 return err;
831}
832
833/*
834 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
835 * looks like the link count will hit 0, drop any other caps (other
836 * than PIN) we don't specifically want (due to the file still being
837 * open).
838 */
839static int drop_caps_for_unlink(struct inode *inode)
840{
841 struct ceph_inode_info *ci = ceph_inode(inode);
842 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
843
Sage Weilbe655592011-11-30 09:47:09 -0800844 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700845 if (inode->i_nlink == 1) {
846 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
847 ci->i_ceph_flags |= CEPH_I_NODELAY;
848 }
Sage Weilbe655592011-11-30 09:47:09 -0800849 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700850 return drop;
851}
852
853/*
854 * rmdir and unlink are differ only by the metadata op code
855 */
856static int ceph_unlink(struct inode *dir, struct dentry *dentry)
857{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700858 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
859 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700860 struct inode *inode = dentry->d_inode;
861 struct ceph_mds_request *req;
862 int err = -EROFS;
863 int op;
864
865 if (ceph_snap(dir) == CEPH_SNAPDIR) {
866 /* rmdir .snap/foo is RMSNAP */
867 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
868 dentry->d_name.name, dentry);
869 op = CEPH_MDS_OP_RMSNAP;
870 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
871 dout("unlink/rmdir dir %p dn %p inode %p\n",
872 dir, dentry, inode);
Al Virodba19c62011-07-25 20:49:29 -0400873 op = S_ISDIR(dentry->d_inode->i_mode) ?
Sage Weil2817b002009-10-06 11:31:08 -0700874 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
875 } else
876 goto out;
877 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
878 if (IS_ERR(req)) {
879 err = PTR_ERR(req);
880 goto out;
881 }
882 req->r_dentry = dget(dentry);
883 req->r_num_caps = 2;
884 req->r_locked_dir = dir;
885 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
886 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
887 req->r_inode_drop = drop_caps_for_unlink(inode);
888 err = ceph_mdsc_do_request(mdsc, dir, req);
889 if (!err && !req->r_reply_info.head->is_dentry)
890 d_delete(dentry);
891 ceph_mdsc_put_request(req);
892out:
893 return err;
894}
895
896static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
897 struct inode *new_dir, struct dentry *new_dentry)
898{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700899 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
900 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700901 struct ceph_mds_request *req;
902 int err;
903
904 if (ceph_snap(old_dir) != ceph_snap(new_dir))
905 return -EXDEV;
906 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
907 ceph_snap(new_dir) != CEPH_NOSNAP)
908 return -EROFS;
909 dout("rename dir %p dentry %p to dir %p dentry %p\n",
910 old_dir, old_dentry, new_dir, new_dentry);
911 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
912 if (IS_ERR(req))
913 return PTR_ERR(req);
914 req->r_dentry = dget(new_dentry);
915 req->r_num_caps = 2;
916 req->r_old_dentry = dget(old_dentry);
Sage Weil41b02e12011-07-26 11:31:14 -0700917 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700918 req->r_locked_dir = new_dir;
919 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
920 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
921 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
922 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
923 /* release LINK_RDCACHE on source inode (mds will lock it) */
924 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
925 if (new_dentry->d_inode)
926 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
927 err = ceph_mdsc_do_request(mdsc, old_dir, req);
928 if (!err && !req->r_reply_info.head->is_dentry) {
929 /*
930 * Normally d_move() is done by fill_trace (called by
931 * do_request, above). If there is no trace, we need
932 * to do it here.
933 */
Sage Weilea1409f2010-04-28 16:12:06 -0700934
935 /* d_move screws up d_subdirs order */
Sage Weilc6ffe102011-11-03 09:23:36 -0700936 ceph_dir_clear_complete(new_dir);
Sage Weilea1409f2010-04-28 16:12:06 -0700937
Sage Weil2817b002009-10-06 11:31:08 -0700938 d_move(old_dentry, new_dentry);
Sage Weilea1409f2010-04-28 16:12:06 -0700939
940 /* ensure target dentry is invalidated, despite
941 rehashing bug in vfs_rename_dir */
Sage Weil81a6cf22010-05-14 09:35:38 -0700942 ceph_invalidate_dentry_lease(new_dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700943 }
944 ceph_mdsc_put_request(req);
945 return err;
946}
947
Sage Weil81a6cf22010-05-14 09:35:38 -0700948/*
949 * Ensure a dentry lease will no longer revalidate.
950 */
951void ceph_invalidate_dentry_lease(struct dentry *dentry)
952{
953 spin_lock(&dentry->d_lock);
954 dentry->d_time = jiffies;
955 ceph_dentry(dentry)->lease_shared_gen = 0;
956 spin_unlock(&dentry->d_lock);
957}
Sage Weil2817b002009-10-06 11:31:08 -0700958
959/*
960 * Check if dentry lease is valid. If not, delete the lease. Try to
961 * renew if the least is more than half up.
962 */
963static int dentry_lease_is_valid(struct dentry *dentry)
964{
965 struct ceph_dentry_info *di;
966 struct ceph_mds_session *s;
967 int valid = 0;
968 u32 gen;
969 unsigned long ttl;
970 struct ceph_mds_session *session = NULL;
971 struct inode *dir = NULL;
972 u32 seq = 0;
973
974 spin_lock(&dentry->d_lock);
975 di = ceph_dentry(dentry);
Sage Weil3d8eb7a2011-11-11 09:48:53 -0800976 if (di->lease_session) {
Sage Weil2817b002009-10-06 11:31:08 -0700977 s = di->lease_session;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800978 spin_lock(&s->s_gen_ttl_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700979 gen = s->s_cap_gen;
980 ttl = s->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800981 spin_unlock(&s->s_gen_ttl_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700982
983 if (di->lease_gen == gen &&
984 time_before(jiffies, dentry->d_time) &&
985 time_before(jiffies, ttl)) {
986 valid = 1;
987 if (di->lease_renew_after &&
988 time_after(jiffies, di->lease_renew_after)) {
989 /* we should renew */
990 dir = dentry->d_parent->d_inode;
991 session = ceph_get_mds_session(s);
992 seq = di->lease_seq;
993 di->lease_renew_after = 0;
994 di->lease_renew_from = jiffies;
995 }
Sage Weil2817b002009-10-06 11:31:08 -0700996 }
997 }
998 spin_unlock(&dentry->d_lock);
999
1000 if (session) {
1001 ceph_mdsc_lease_send_msg(session, dir, dentry,
1002 CEPH_MDS_LEASE_RENEW, seq);
1003 ceph_put_mds_session(session);
1004 }
1005 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1006 return valid;
1007}
1008
1009/*
1010 * Check if directory-wide content lease/cap is valid.
1011 */
1012static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1013{
1014 struct ceph_inode_info *ci = ceph_inode(dir);
1015 struct ceph_dentry_info *di = ceph_dentry(dentry);
1016 int valid = 0;
1017
Sage Weilbe655592011-11-30 09:47:09 -08001018 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001019 if (ci->i_shared_gen == di->lease_shared_gen)
1020 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
Sage Weilbe655592011-11-30 09:47:09 -08001021 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001022 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1023 dir, (unsigned)ci->i_shared_gen, dentry,
1024 (unsigned)di->lease_shared_gen, valid);
1025 return valid;
1026}
1027
1028/*
1029 * Check if cached dentry can be trusted.
1030 */
1031static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
1032{
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001033 int valid = 0;
Nick Piggin34286d62011-01-07 17:49:57 +11001034 struct inode *dir;
1035
Al Viro0eb980e2011-03-10 03:44:05 -05001036 if (nd && nd->flags & LOOKUP_RCU)
Nick Piggin34286d62011-01-07 17:49:57 +11001037 return -ECHILD;
1038
Sage Weil1cd39352010-05-03 22:08:02 -07001039 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1040 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1041 ceph_dentry(dentry)->offset);
Sage Weil2817b002009-10-06 11:31:08 -07001042
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001043 dir = ceph_get_dentry_parent_inode(dentry);
1044
Sage Weil2817b002009-10-06 11:31:08 -07001045 /* always trust cached snapped dentries, snapdir dentry */
1046 if (ceph_snap(dir) != CEPH_NOSNAP) {
1047 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1048 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001049 valid = 1;
1050 } else if (dentry->d_inode &&
1051 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1052 valid = 1;
1053 } else if (dentry_lease_is_valid(dentry) ||
1054 dir_lease_is_valid(dir, dentry)) {
1055 valid = 1;
Sage Weil2817b002009-10-06 11:31:08 -07001056 }
Sage Weil2817b002009-10-06 11:31:08 -07001057
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001058 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1059 if (valid)
1060 ceph_dentry_lru_touch(dentry);
1061 else
1062 d_drop(dentry);
1063 iput(dir);
1064 return valid;
Sage Weil2817b002009-10-06 11:31:08 -07001065}
1066
1067/*
Sage Weil147851d2011-03-15 14:57:41 -07001068 * Release our ceph_dentry_info.
Sage Weil2817b002009-10-06 11:31:08 -07001069 */
Sage Weil147851d2011-03-15 14:57:41 -07001070static void ceph_d_release(struct dentry *dentry)
Sage Weil2817b002009-10-06 11:31:08 -07001071{
1072 struct ceph_dentry_info *di = ceph_dentry(dentry);
Sage Weil2817b002009-10-06 11:31:08 -07001073
Sage Weil147851d2011-03-15 14:57:41 -07001074 dout("d_release %p\n", dentry);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001075 ceph_dentry_lru_del(dentry);
1076 if (di->lease_session)
1077 ceph_put_mds_session(di->lease_session);
1078 kmem_cache_free(ceph_dentry_cachep, di);
1079 dentry->d_fsdata = NULL;
Sage Weil2817b002009-10-06 11:31:08 -07001080}
1081
1082static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1083 struct nameidata *nd)
1084{
1085 /*
1086 * Eventually, we'll want to revalidate snapped metadata
1087 * too... probably...
1088 */
1089 return 1;
1090}
1091
Sage Weilb58dc412011-03-15 15:53:40 -07001092/*
Sage Weilc6ffe102011-11-03 09:23:36 -07001093 * Set/clear/test dir complete flag on the dir's dentry.
1094 */
Sage Weilc6ffe102011-11-03 09:23:36 -07001095void ceph_dir_set_complete(struct inode *inode)
1096{
Sage Weila40dc6c2012-01-10 09:12:55 -08001097 struct dentry *dentry = d_find_any_alias(inode);
1098
1099 if (dentry && ceph_dentry(dentry) &&
1100 ceph_test_mount_opt(ceph_sb_to_client(dentry->d_sb), DCACHE)) {
1101 dout(" marking %p (%p) complete\n", inode, dentry);
1102 set_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1103 }
1104 dput(dentry);
Sage Weilc6ffe102011-11-03 09:23:36 -07001105}
1106
1107void ceph_dir_clear_complete(struct inode *inode)
1108{
Sage Weila40dc6c2012-01-10 09:12:55 -08001109 struct dentry *dentry = d_find_any_alias(inode);
1110
1111 if (dentry && ceph_dentry(dentry)) {
1112 dout(" marking %p (%p) complete\n", inode, dentry);
1113 set_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1114 }
1115 dput(dentry);
Sage Weilc6ffe102011-11-03 09:23:36 -07001116}
1117
1118bool ceph_dir_test_complete(struct inode *inode)
1119{
Sage Weila40dc6c2012-01-10 09:12:55 -08001120 struct dentry *dentry = d_find_any_alias(inode);
1121
1122 if (dentry && ceph_dentry(dentry)) {
1123 dout(" marking %p (%p) NOT complete\n", inode, dentry);
1124 clear_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1125 }
1126 dput(dentry);
Sage Weilc6ffe102011-11-03 09:23:36 -07001127 return false;
1128}
1129
1130/*
Sage Weilb58dc412011-03-15 15:53:40 -07001131 * When the VFS prunes a dentry from the cache, we need to clear the
1132 * complete flag on the parent directory.
1133 *
1134 * Called under dentry->d_lock.
1135 */
1136static void ceph_d_prune(struct dentry *dentry)
1137{
1138 struct ceph_dentry_info *di;
Sage Weil2817b002009-10-06 11:31:08 -07001139
Sage Weil774ac212011-11-11 09:48:08 -08001140 dout("ceph_d_prune %p\n", dentry);
Sage Weilb58dc412011-03-15 15:53:40 -07001141
1142 /* do we have a valid parent? */
1143 if (!dentry->d_parent || IS_ROOT(dentry))
1144 return;
1145
1146 /* if we are not hashed, we don't affect D_COMPLETE */
1147 if (d_unhashed(dentry))
1148 return;
1149
1150 /*
1151 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1152 * cleared until d_release
1153 */
1154 di = ceph_dentry(dentry->d_parent);
1155 clear_bit(CEPH_D_COMPLETE, &di->flags);
1156}
Sage Weil2817b002009-10-06 11:31:08 -07001157
1158/*
1159 * read() on a dir. This weird interface hack only works if mounted
1160 * with '-o dirstat'.
1161 */
1162static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1163 loff_t *ppos)
1164{
1165 struct ceph_file_info *cf = file->private_data;
1166 struct inode *inode = file->f_dentry->d_inode;
1167 struct ceph_inode_info *ci = ceph_inode(inode);
1168 int left;
Sage Weilae598082011-05-12 14:28:05 -07001169 const int bufsize = 1024;
Sage Weil2817b002009-10-06 11:31:08 -07001170
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001171 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
Sage Weil2817b002009-10-06 11:31:08 -07001172 return -EISDIR;
1173
1174 if (!cf->dir_info) {
Sage Weilae598082011-05-12 14:28:05 -07001175 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
Sage Weil2817b002009-10-06 11:31:08 -07001176 if (!cf->dir_info)
1177 return -ENOMEM;
1178 cf->dir_info_len =
Sage Weilae598082011-05-12 14:28:05 -07001179 snprintf(cf->dir_info, bufsize,
Sage Weil2817b002009-10-06 11:31:08 -07001180 "entries: %20lld\n"
1181 " files: %20lld\n"
1182 " subdirs: %20lld\n"
1183 "rentries: %20lld\n"
1184 " rfiles: %20lld\n"
1185 " rsubdirs: %20lld\n"
1186 "rbytes: %20lld\n"
1187 "rctime: %10ld.%09ld\n",
1188 ci->i_files + ci->i_subdirs,
1189 ci->i_files,
1190 ci->i_subdirs,
1191 ci->i_rfiles + ci->i_rsubdirs,
1192 ci->i_rfiles,
1193 ci->i_rsubdirs,
1194 ci->i_rbytes,
1195 (long)ci->i_rctime.tv_sec,
1196 (long)ci->i_rctime.tv_nsec);
1197 }
1198
1199 if (*ppos >= cf->dir_info_len)
1200 return 0;
1201 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1202 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1203 if (left == size)
1204 return -EFAULT;
1205 *ppos += (size - left);
1206 return size - left;
1207}
1208
1209/*
1210 * an fsync() on a dir will wait for any uncommitted directory
1211 * operations to commit.
1212 */
Josef Bacik02c24a82011-07-16 20:44:56 -04001213static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1214 int datasync)
Sage Weil2817b002009-10-06 11:31:08 -07001215{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001216 struct inode *inode = file->f_path.dentry->d_inode;
Sage Weil2817b002009-10-06 11:31:08 -07001217 struct ceph_inode_info *ci = ceph_inode(inode);
1218 struct list_head *head = &ci->i_unsafe_dirops;
1219 struct ceph_mds_request *req;
1220 u64 last_tid;
1221 int ret = 0;
1222
1223 dout("dir_fsync %p\n", inode);
Josef Bacik02c24a82011-07-16 20:44:56 -04001224 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1225 if (ret)
1226 return ret;
1227 mutex_lock(&inode->i_mutex);
1228
Sage Weil2817b002009-10-06 11:31:08 -07001229 spin_lock(&ci->i_unsafe_lock);
1230 if (list_empty(head))
1231 goto out;
1232
1233 req = list_entry(head->prev,
1234 struct ceph_mds_request, r_unsafe_dir_item);
1235 last_tid = req->r_tid;
1236
1237 do {
1238 ceph_mdsc_get_request(req);
1239 spin_unlock(&ci->i_unsafe_lock);
Sage Weil2ff179e2012-01-03 10:09:07 -08001240
Sage Weil2817b002009-10-06 11:31:08 -07001241 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1242 inode, req->r_tid, last_tid);
1243 if (req->r_timeout) {
1244 ret = wait_for_completion_timeout(
1245 &req->r_safe_completion, req->r_timeout);
1246 if (ret > 0)
1247 ret = 0;
1248 else if (ret == 0)
1249 ret = -EIO; /* timed out */
1250 } else {
1251 wait_for_completion(&req->r_safe_completion);
1252 }
Sage Weil2817b002009-10-06 11:31:08 -07001253 ceph_mdsc_put_request(req);
1254
Sage Weil2ff179e2012-01-03 10:09:07 -08001255 spin_lock(&ci->i_unsafe_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001256 if (ret || list_empty(head))
1257 break;
1258 req = list_entry(head->next,
1259 struct ceph_mds_request, r_unsafe_dir_item);
1260 } while (req->r_tid < last_tid);
1261out:
1262 spin_unlock(&ci->i_unsafe_lock);
Josef Bacik02c24a82011-07-16 20:44:56 -04001263 mutex_unlock(&inode->i_mutex);
1264
Sage Weil2817b002009-10-06 11:31:08 -07001265 return ret;
1266}
1267
1268/*
1269 * We maintain a private dentry LRU.
1270 *
1271 * FIXME: this needs to be changed to a per-mds lru to be useful.
1272 */
1273void ceph_dentry_lru_add(struct dentry *dn)
1274{
1275 struct ceph_dentry_info *di = ceph_dentry(dn);
1276 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001277
Sage Weil04a419f2009-12-23 09:30:21 -08001278 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1279 dn->d_name.len, dn->d_name.name);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001280 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1281 spin_lock(&mdsc->dentry_lru_lock);
1282 list_add_tail(&di->lru, &mdsc->dentry_lru);
1283 mdsc->num_dentry++;
1284 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001285}
1286
1287void ceph_dentry_lru_touch(struct dentry *dn)
1288{
1289 struct ceph_dentry_info *di = ceph_dentry(dn);
1290 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001291
Sage Weil1cd39352010-05-03 22:08:02 -07001292 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1293 dn->d_name.len, dn->d_name.name, di->offset);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001294 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1295 spin_lock(&mdsc->dentry_lru_lock);
1296 list_move_tail(&di->lru, &mdsc->dentry_lru);
1297 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001298}
1299
1300void ceph_dentry_lru_del(struct dentry *dn)
1301{
1302 struct ceph_dentry_info *di = ceph_dentry(dn);
1303 struct ceph_mds_client *mdsc;
1304
Sage Weil04a419f2009-12-23 09:30:21 -08001305 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1306 dn->d_name.len, dn->d_name.name);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001307 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1308 spin_lock(&mdsc->dentry_lru_lock);
1309 list_del_init(&di->lru);
1310 mdsc->num_dentry--;
1311 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001312}
1313
Sage Weil6c0f3af2010-11-16 11:14:34 -08001314/*
1315 * Return name hash for a given dentry. This is dependent on
1316 * the parent directory's hash function.
1317 */
Sage Weile5f86dc2011-07-26 11:30:55 -07001318unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
Sage Weil6c0f3af2010-11-16 11:14:34 -08001319{
Sage Weil6c0f3af2010-11-16 11:14:34 -08001320 struct ceph_inode_info *dci = ceph_inode(dir);
1321
1322 switch (dci->i_dir_layout.dl_dir_hash) {
1323 case 0: /* for backward compat */
1324 case CEPH_STR_HASH_LINUX:
1325 return dn->d_name.hash;
1326
1327 default:
1328 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1329 dn->d_name.name, dn->d_name.len);
1330 }
1331}
1332
Sage Weil2817b002009-10-06 11:31:08 -07001333const struct file_operations ceph_dir_fops = {
1334 .read = ceph_read_dir,
1335 .readdir = ceph_readdir,
1336 .llseek = ceph_dir_llseek,
1337 .open = ceph_open,
1338 .release = ceph_release,
1339 .unlocked_ioctl = ceph_ioctl,
1340 .fsync = ceph_dir_fsync,
1341};
1342
1343const struct inode_operations ceph_dir_iops = {
1344 .lookup = ceph_lookup,
1345 .permission = ceph_permission,
1346 .getattr = ceph_getattr,
1347 .setattr = ceph_setattr,
1348 .setxattr = ceph_setxattr,
1349 .getxattr = ceph_getxattr,
1350 .listxattr = ceph_listxattr,
1351 .removexattr = ceph_removexattr,
1352 .mknod = ceph_mknod,
1353 .symlink = ceph_symlink,
1354 .mkdir = ceph_mkdir,
1355 .link = ceph_link,
1356 .unlink = ceph_unlink,
1357 .rmdir = ceph_unlink,
1358 .rename = ceph_rename,
1359 .create = ceph_create,
1360};
1361
Sage Weil52dfb8a2010-08-03 10:25:30 -07001362const struct dentry_operations ceph_dentry_ops = {
Sage Weil2817b002009-10-06 11:31:08 -07001363 .d_revalidate = ceph_d_revalidate,
Sage Weil147851d2011-03-15 14:57:41 -07001364 .d_release = ceph_d_release,
Sage Weilb58dc412011-03-15 15:53:40 -07001365 .d_prune = ceph_d_prune,
Sage Weil2817b002009-10-06 11:31:08 -07001366};
1367
Sage Weil52dfb8a2010-08-03 10:25:30 -07001368const struct dentry_operations ceph_snapdir_dentry_ops = {
Sage Weil2817b002009-10-06 11:31:08 -07001369 .d_revalidate = ceph_snapdir_d_revalidate,
Sage Weil147851d2011-03-15 14:57:41 -07001370 .d_release = ceph_d_release,
Sage Weil2817b002009-10-06 11:31:08 -07001371};
1372
Sage Weil52dfb8a2010-08-03 10:25:30 -07001373const struct dentry_operations ceph_snap_dentry_ops = {
Sage Weil147851d2011-03-15 14:57:41 -07001374 .d_release = ceph_d_release,
Sage Weilb58dc412011-03-15 15:53:40 -07001375 .d_prune = ceph_d_prune,
Sage Weil2817b002009-10-06 11:31:08 -07001376};