]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
2817b000 SW |
2 | |
3 | #include <linux/spinlock.h> | |
4 | #include <linux/fs_struct.h> | |
5 | #include <linux/namei.h> | |
5a0e3ad6 | 6 | #include <linux/slab.h> |
2817b000 SW |
7 | #include <linux/sched.h> |
8 | ||
9 | #include "super.h" | |
3d14c5d2 | 10 | #include "mds_client.h" |
2817b000 SW |
11 | |
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 | ||
29 | const struct inode_operations ceph_dir_iops; | |
30 | const struct file_operations ceph_dir_fops; | |
52dfb8ac | 31 | const struct dentry_operations ceph_dentry_ops; |
2817b000 SW |
32 | |
33 | /* | |
34 | * Initialize ceph dentry state. | |
35 | */ | |
36 | int ceph_init_dentry(struct dentry *dentry) | |
37 | { | |
38 | struct ceph_dentry_info *di; | |
39 | ||
40 | if (dentry->d_fsdata) | |
41 | return 0; | |
42 | ||
92cf7652 SW |
43 | if (dentry->d_parent == NULL || /* nfs fh_to_dentry */ |
44 | ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) | |
fb045adb | 45 | d_set_d_op(dentry, &ceph_dentry_ops); |
2817b000 | 46 | else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR) |
fb045adb | 47 | d_set_d_op(dentry, &ceph_snapdir_dentry_ops); |
2817b000 | 48 | else |
fb045adb | 49 | d_set_d_op(dentry, &ceph_snap_dentry_ops); |
2817b000 | 50 | |
36e21687 | 51 | di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO); |
2817b000 SW |
52 | if (!di) |
53 | return -ENOMEM; /* oh well */ | |
54 | ||
55 | spin_lock(&dentry->d_lock); | |
8c6efb58 SW |
56 | if (dentry->d_fsdata) { |
57 | /* lost a race */ | |
58 | kmem_cache_free(ceph_dentry_cachep, di); | |
2817b000 | 59 | goto out_unlock; |
8c6efb58 | 60 | } |
2817b000 SW |
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; | |
69 | } | |
70 | ||
71 | ||
72 | ||
73 | /* | |
74 | * for readdir, we encode the directory frag and offset within that | |
75 | * frag into f_pos. | |
76 | */ | |
77 | static unsigned fpos_frag(loff_t p) | |
78 | { | |
79 | return p >> 32; | |
80 | } | |
81 | static unsigned fpos_off(loff_t p) | |
82 | { | |
83 | return p & 0xffffffff; | |
84 | } | |
85 | ||
86 | /* | |
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. | |
92 | * | |
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). | |
96 | */ | |
97 | static int __dcache_readdir(struct file *filp, | |
98 | void *dirent, filldir_t filldir) | |
99 | { | |
2817b000 SW |
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; | |
107 | ||
108 | /* claim ref on last dentry we returned */ | |
109 | last = fi->dentry; | |
110 | fi->dentry = NULL; | |
111 | ||
112 | dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos, | |
113 | last); | |
114 | ||
2fd6b7f5 | 115 | spin_lock(&parent->d_lock); |
2817b000 SW |
116 | |
117 | /* start at beginning? */ | |
884ea892 SW |
118 | if (filp->f_pos == 2 || last == NULL || |
119 | filp->f_pos < ceph_dentry(last)->offset) { | |
2817b000 SW |
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; | |
126 | } | |
127 | ||
128 | more: | |
129 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
130 | di = ceph_dentry(dentry); | |
131 | while (1) { | |
1cd3935b SW |
132 | dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next, |
133 | d_unhashed(dentry) ? "!hashed" : "hashed", | |
2817b000 SW |
134 | parent->d_subdirs.prev, parent->d_subdirs.next); |
135 | if (p == &parent->d_subdirs) { | |
9cfa1098 | 136 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
137 | goto out_unlock; |
138 | } | |
2fd6b7f5 | 139 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2817b000 | 140 | if (!d_unhashed(dentry) && dentry->d_inode && |
09b8a7d2 | 141 | ceph_snap(dentry->d_inode) != CEPH_SNAPDIR && |
1d1de916 | 142 | ceph_ino(dentry->d_inode) != CEPH_INO_CEPH && |
2817b000 SW |
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" : ""); | |
da502956 | 149 | spin_unlock(&dentry->d_lock); |
2817b000 SW |
150 | p = p->prev; |
151 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
152 | di = ceph_dentry(dentry); | |
153 | } | |
154 | ||
da502956 | 155 | dget_dlock(dentry); |
b7ab39f6 | 156 | spin_unlock(&dentry->d_lock); |
2fd6b7f5 | 157 | spin_unlock(&parent->d_lock); |
2817b000 SW |
158 | |
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, | |
ad1fee96 | 164 | ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino), |
2817b000 SW |
165 | dentry->d_inode->i_mode >> 12); |
166 | ||
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); | |
174 | } | |
2817b000 | 175 | } |
f5b06628 SW |
176 | last = dentry; |
177 | ||
2817b000 | 178 | if (err < 0) |
efa4c120 | 179 | goto out; |
2817b000 | 180 | |
2817b000 SW |
181 | filp->f_pos++; |
182 | ||
b5c84bf6 | 183 | /* make sure a dentry wasn't dropped while we didn't have parent lock */ |
efa4c120 SW |
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; | |
188 | } | |
189 | ||
2fd6b7f5 | 190 | spin_lock(&parent->d_lock); |
efa4c120 SW |
191 | p = p->prev; /* advance to next dentry */ |
192 | goto more; | |
2817b000 SW |
193 | |
194 | out_unlock: | |
2fd6b7f5 | 195 | spin_unlock(&parent->d_lock); |
efa4c120 SW |
196 | out: |
197 | if (last) | |
2817b000 | 198 | dput(last); |
2817b000 SW |
199 | return err; |
200 | } | |
201 | ||
202 | /* | |
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. | |
207 | */ | |
208 | static int note_last_dentry(struct ceph_file_info *fi, const char *name, | |
209 | int len) | |
210 | { | |
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; | |
219 | } | |
220 | ||
221 | static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
222 | { | |
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); | |
3d14c5d2 YS |
226 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
227 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
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; | |
3d14c5d2 YS |
233 | const int max_entries = fsc->mount_options->max_readdir; |
234 | const int max_bytes = fsc->mount_options->max_readdir_bytes; | |
2817b000 SW |
235 | |
236 | dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off); | |
9cfa1098 | 237 | if (fi->flags & CEPH_F_ATEND) |
2817b000 SW |
238 | return 0; |
239 | ||
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; | |
245 | ||
246 | dout("readdir off 0 -> '.'\n"); | |
247 | if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0), | |
ad1fee96 YS |
248 | ceph_translate_ino(inode->i_sb, inode->i_ino), |
249 | inode->i_mode >> 12) < 0) | |
2817b000 SW |
250 | return 0; |
251 | filp->f_pos = 1; | |
252 | off = 1; | |
253 | } | |
254 | if (filp->f_pos == 1) { | |
ad1fee96 | 255 | ino_t ino = filp->f_dentry->d_parent->d_inode->i_ino; |
2817b000 SW |
256 | dout("readdir off 1 -> '..'\n"); |
257 | if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1), | |
ad1fee96 | 258 | ceph_translate_ino(inode->i_sb, ino), |
2817b000 SW |
259 | inode->i_mode >> 12) < 0) |
260 | return 0; | |
261 | filp->f_pos = 2; | |
262 | off = 2; | |
263 | } | |
264 | ||
265 | /* can we use the dcache? */ | |
266 | spin_lock(&inode->i_lock); | |
267 | if ((filp->f_pos == 2 || fi->dentry) && | |
3d14c5d2 | 268 | !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && |
a0dff78d | 269 | ceph_snap(inode) != CEPH_SNAPDIR && |
2817b000 SW |
270 | (ci->i_ceph_flags & CEPH_I_COMPLETE) && |
271 | __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { | |
efa4c120 | 272 | spin_unlock(&inode->i_lock); |
2817b000 | 273 | err = __dcache_readdir(filp, dirent, filldir); |
efa4c120 | 274 | if (err != -EAGAIN) |
2817b000 | 275 | return err; |
efa4c120 SW |
276 | } else { |
277 | spin_unlock(&inode->i_lock); | |
2817b000 | 278 | } |
2817b000 SW |
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; | |
286 | } | |
287 | ||
288 | /* proceed with a normal readdir */ | |
289 | ||
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; | |
296 | ||
297 | /* discard old result, if any */ | |
393f6620 | 298 | if (fi->last_readdir) { |
2817b000 | 299 | ceph_mdsc_put_request(fi->last_readdir); |
393f6620 SW |
300 | fi->last_readdir = NULL; |
301 | } | |
2817b000 SW |
302 | |
303 | /* requery frag tree, as the frag topology may have changed */ | |
304 | frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL); | |
305 | ||
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); | |
70b666c3 SW |
311 | req->r_inode = inode; |
312 | ihold(inode); | |
2817b000 SW |
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); | |
23804d91 | 322 | req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes); |
e1e4dd0c | 323 | req->r_num_caps = max_entries + 1; |
2817b000 SW |
324 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
325 | if (err < 0) { | |
326 | ceph_mdsc_put_request(req); | |
327 | return err; | |
328 | } | |
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); | |
333 | ||
334 | if (!req->r_did_prepopulate) { | |
335 | dout("readdir !did_prepopulate"); | |
336 | fi->dir_release_count--; /* preclude I_COMPLETE */ | |
337 | } | |
338 | ||
339 | /* note next offset and last dentry name */ | |
340 | fi->offset = fi->next_offset; | |
341 | fi->last_readdir = req; | |
342 | ||
343 | if (req->r_reply_info.dir_end) { | |
344 | kfree(fi->last_name); | |
345 | fi->last_name = NULL; | |
7b88dadc SW |
346 | if (ceph_frag_is_rightmost(frag)) |
347 | fi->next_offset = 2; | |
348 | else | |
349 | fi->next_offset = 0; | |
2817b000 SW |
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; | |
358 | } | |
359 | } | |
360 | ||
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); | |
da39822c | 364 | while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) { |
2817b000 SW |
365 | u64 pos = ceph_make_fpos(frag, off); |
366 | struct ceph_mds_reply_inode *in = | |
367 | rinfo->dir_in[off - fi->offset].in; | |
3105c19c SW |
368 | struct ceph_vino vino; |
369 | ino_t ino; | |
370 | ||
2817b000 SW |
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; | |
3105c19c SW |
377 | vino.ino = le64_to_cpu(in->ino); |
378 | vino.snap = le64_to_cpu(in->snapid); | |
379 | ino = ceph_vino_to_ino(vino); | |
2817b000 SW |
380 | if (filldir(dirent, |
381 | rinfo->dir_dname[off - fi->offset], | |
382 | rinfo->dir_dname_len[off - fi->offset], | |
ad1fee96 YS |
383 | pos, |
384 | ceph_translate_ino(inode->i_sb, ino), ftype) < 0) { | |
2817b000 SW |
385 | dout("filldir stopping us...\n"); |
386 | return 0; | |
387 | } | |
388 | off++; | |
389 | filp->f_pos = pos + 1; | |
390 | } | |
391 | ||
392 | if (fi->last_name) { | |
393 | ceph_mdsc_put_request(fi->last_readdir); | |
394 | fi->last_readdir = NULL; | |
395 | goto more; | |
396 | } | |
397 | ||
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; | |
405 | } | |
9cfa1098 | 406 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
407 | |
408 | /* | |
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. | |
412 | */ | |
413 | spin_lock(&inode->i_lock); | |
414 | if (ci->i_release_count == fi->dir_release_count) { | |
415 | dout(" marking %p complete\n", inode); | |
b545cc15 | 416 | /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */ |
2817b000 SW |
417 | ci->i_max_offset = filp->f_pos; |
418 | } | |
419 | spin_unlock(&inode->i_lock); | |
420 | ||
421 | dout("readdir %p filp %p done.\n", inode, filp); | |
422 | return 0; | |
423 | } | |
424 | ||
425 | static void reset_readdir(struct ceph_file_info *fi) | |
426 | { | |
427 | if (fi->last_readdir) { | |
428 | ceph_mdsc_put_request(fi->last_readdir); | |
429 | fi->last_readdir = NULL; | |
430 | } | |
431 | kfree(fi->last_name); | |
a1629c3b | 432 | fi->last_name = NULL; |
2817b000 SW |
433 | fi->next_offset = 2; /* compensate for . and .. */ |
434 | if (fi->dentry) { | |
435 | dput(fi->dentry); | |
436 | fi->dentry = NULL; | |
437 | } | |
9cfa1098 | 438 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
439 | } |
440 | ||
441 | static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin) | |
442 | { | |
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; | |
447 | ||
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; | |
455 | } | |
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; | |
9cfa1098 | 461 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
462 | } |
463 | retval = offset; | |
464 | ||
465 | /* | |
466 | * discard buffered readdir content on seekdir(0), or | |
467 | * seek to new frag, or seek prior to current chunk. | |
468 | */ | |
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); | |
474 | } | |
475 | ||
476 | /* bump dir_release_count if we did a forward seek */ | |
477 | if (offset > old_offset) | |
478 | fi->dir_release_count--; | |
479 | } | |
480 | mutex_unlock(&inode->i_mutex); | |
481 | return retval; | |
482 | } | |
483 | ||
484 | /* | |
468640e3 | 485 | * Handle lookups for the hidden .snap directory. |
2817b000 | 486 | */ |
468640e3 SW |
487 | int ceph_handle_snapdir(struct ceph_mds_request *req, |
488 | struct dentry *dentry, int err) | |
2817b000 | 489 | { |
3d14c5d2 | 490 | struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); |
2817b000 SW |
491 | struct inode *parent = dentry->d_parent->d_inode; |
492 | ||
493 | /* .snap dir? */ | |
494 | if (err == -ENOENT && | |
455cec0a | 495 | ceph_snap(parent) == CEPH_NOSNAP && |
6b805185 | 496 | strcmp(dentry->d_name.name, |
3d14c5d2 | 497 | fsc->mount_options->snapdir_name) == 0) { |
2817b000 SW |
498 | struct inode *inode = ceph_get_snapdir(parent); |
499 | dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n", | |
500 | dentry, dentry->d_name.len, dentry->d_name.name, inode); | |
9358c6d4 | 501 | BUG_ON(!d_unhashed(dentry)); |
2817b000 SW |
502 | d_add(dentry, inode); |
503 | err = 0; | |
504 | } | |
468640e3 SW |
505 | return err; |
506 | } | |
2817b000 | 507 | |
468640e3 SW |
508 | /* |
509 | * Figure out final result of a lookup/open request. | |
510 | * | |
511 | * Mainly, make sure we return the final req->r_dentry (if it already | |
512 | * existed) in place of the original VFS-provided dentry when they | |
513 | * differ. | |
514 | * | |
515 | * Gracefully handle the case where the MDS replies with -ENOENT and | |
516 | * no trace (which it may do, at its discretion, e.g., if it doesn't | |
517 | * care to issue a lease on the negative dentry). | |
518 | */ | |
519 | struct dentry *ceph_finish_lookup(struct ceph_mds_request *req, | |
520 | struct dentry *dentry, int err) | |
521 | { | |
2817b000 SW |
522 | if (err == -ENOENT) { |
523 | /* no trace? */ | |
524 | err = 0; | |
525 | if (!req->r_reply_info.head->is_dentry) { | |
526 | dout("ENOENT and no trace, dentry %p inode %p\n", | |
527 | dentry, dentry->d_inode); | |
528 | if (dentry->d_inode) { | |
529 | d_drop(dentry); | |
530 | err = -ENOENT; | |
531 | } else { | |
532 | d_add(dentry, NULL); | |
533 | } | |
534 | } | |
535 | } | |
536 | if (err) | |
537 | dentry = ERR_PTR(err); | |
538 | else if (dentry != req->r_dentry) | |
539 | dentry = dget(req->r_dentry); /* we got spliced */ | |
540 | else | |
541 | dentry = NULL; | |
542 | return dentry; | |
543 | } | |
544 | ||
1d1de916 SW |
545 | static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry) |
546 | { | |
547 | return ceph_ino(inode) == CEPH_INO_ROOT && | |
548 | strncmp(dentry->d_name.name, ".ceph", 5) == 0; | |
549 | } | |
550 | ||
2817b000 SW |
551 | /* |
552 | * Look up a single dir entry. If there is a lookup intent, inform | |
553 | * the MDS so that it gets our 'caps wanted' value in a single op. | |
554 | */ | |
555 | static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, | |
556 | struct nameidata *nd) | |
557 | { | |
3d14c5d2 YS |
558 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
559 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
560 | struct ceph_mds_request *req; |
561 | int op; | |
562 | int err; | |
563 | ||
564 | dout("lookup %p dentry %p '%.*s'\n", | |
565 | dir, dentry, dentry->d_name.len, dentry->d_name.name); | |
566 | ||
567 | if (dentry->d_name.len > NAME_MAX) | |
568 | return ERR_PTR(-ENAMETOOLONG); | |
569 | ||
570 | err = ceph_init_dentry(dentry); | |
571 | if (err < 0) | |
572 | return ERR_PTR(err); | |
573 | ||
574 | /* open (but not create!) intent? */ | |
575 | if (nd && | |
576 | (nd->flags & LOOKUP_OPEN) && | |
577 | (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */ | |
578 | !(nd->intent.open.flags & O_CREAT)) { | |
579 | int mode = nd->intent.open.create_mode & ~current->fs->umask; | |
580 | return ceph_lookup_open(dir, dentry, nd, mode, 1); | |
581 | } | |
582 | ||
583 | /* can we conclude ENOENT locally? */ | |
584 | if (dentry->d_inode == NULL) { | |
585 | struct ceph_inode_info *ci = ceph_inode(dir); | |
586 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
587 | ||
588 | spin_lock(&dir->i_lock); | |
589 | dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags); | |
590 | if (strncmp(dentry->d_name.name, | |
3d14c5d2 | 591 | fsc->mount_options->snapdir_name, |
2817b000 | 592 | dentry->d_name.len) && |
1d1de916 | 593 | !is_root_ceph_dentry(dir, dentry) && |
2817b000 SW |
594 | (ci->i_ceph_flags & CEPH_I_COMPLETE) && |
595 | (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) { | |
2817b000 SW |
596 | spin_unlock(&dir->i_lock); |
597 | dout(" dir %p complete, -ENOENT\n", dir); | |
598 | d_add(dentry, NULL); | |
599 | di->lease_shared_gen = ci->i_shared_gen; | |
600 | return NULL; | |
601 | } | |
602 | spin_unlock(&dir->i_lock); | |
603 | } | |
604 | ||
605 | op = ceph_snap(dir) == CEPH_SNAPDIR ? | |
606 | CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
607 | req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); | |
608 | if (IS_ERR(req)) | |
7e34bc52 | 609 | return ERR_CAST(req); |
2817b000 SW |
610 | req->r_dentry = dget(dentry); |
611 | req->r_num_caps = 2; | |
612 | /* we only need inode linkage */ | |
613 | req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE); | |
614 | req->r_locked_dir = dir; | |
615 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
468640e3 | 616 | err = ceph_handle_snapdir(req, dentry, err); |
2817b000 SW |
617 | dentry = ceph_finish_lookup(req, dentry, err); |
618 | ceph_mdsc_put_request(req); /* will dput(dentry) */ | |
619 | dout("lookup result=%p\n", dentry); | |
620 | return dentry; | |
621 | } | |
622 | ||
623 | /* | |
624 | * If we do a create but get no trace back from the MDS, follow up with | |
625 | * a lookup (the VFS expects us to link up the provided dentry). | |
626 | */ | |
627 | int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry) | |
628 | { | |
629 | struct dentry *result = ceph_lookup(dir, dentry, NULL); | |
630 | ||
631 | if (result && !IS_ERR(result)) { | |
632 | /* | |
633 | * We created the item, then did a lookup, and found | |
634 | * it was already linked to another inode we already | |
635 | * had in our cache (and thus got spliced). Link our | |
636 | * dentry to that inode, but don't hash it, just in | |
637 | * case the VFS wants to dereference it. | |
638 | */ | |
639 | BUG_ON(!result->d_inode); | |
640 | d_instantiate(dentry, result->d_inode); | |
641 | return 0; | |
642 | } | |
643 | return PTR_ERR(result); | |
644 | } | |
645 | ||
646 | static int ceph_mknod(struct inode *dir, struct dentry *dentry, | |
647 | int mode, dev_t rdev) | |
648 | { | |
3d14c5d2 YS |
649 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
650 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
651 | struct ceph_mds_request *req; |
652 | int err; | |
653 | ||
654 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
655 | return -EROFS; | |
656 | ||
657 | dout("mknod in dir %p dentry %p mode 0%o rdev %d\n", | |
658 | dir, dentry, mode, rdev); | |
659 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS); | |
660 | if (IS_ERR(req)) { | |
661 | d_drop(dentry); | |
662 | return PTR_ERR(req); | |
663 | } | |
664 | req->r_dentry = dget(dentry); | |
665 | req->r_num_caps = 2; | |
666 | req->r_locked_dir = dir; | |
667 | req->r_args.mknod.mode = cpu_to_le32(mode); | |
668 | req->r_args.mknod.rdev = cpu_to_le32(rdev); | |
669 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
670 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
671 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
672 | if (!err && !req->r_reply_info.head->is_dentry) | |
673 | err = ceph_handle_notrace_create(dir, dentry); | |
674 | ceph_mdsc_put_request(req); | |
675 | if (err) | |
676 | d_drop(dentry); | |
677 | return err; | |
678 | } | |
679 | ||
680 | static int ceph_create(struct inode *dir, struct dentry *dentry, int mode, | |
681 | struct nameidata *nd) | |
682 | { | |
683 | dout("create in dir %p dentry %p name '%.*s'\n", | |
684 | dir, dentry, dentry->d_name.len, dentry->d_name.name); | |
685 | ||
686 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
687 | return -EROFS; | |
688 | ||
689 | if (nd) { | |
690 | BUG_ON((nd->flags & LOOKUP_OPEN) == 0); | |
691 | dentry = ceph_lookup_open(dir, dentry, nd, mode, 0); | |
692 | /* hrm, what should i do here if we get aliased? */ | |
693 | if (IS_ERR(dentry)) | |
694 | return PTR_ERR(dentry); | |
695 | return 0; | |
696 | } | |
697 | ||
698 | /* fall back to mknod */ | |
699 | return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0); | |
700 | } | |
701 | ||
702 | static int ceph_symlink(struct inode *dir, struct dentry *dentry, | |
703 | const char *dest) | |
704 | { | |
3d14c5d2 YS |
705 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
706 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
707 | struct ceph_mds_request *req; |
708 | int err; | |
709 | ||
710 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
711 | return -EROFS; | |
712 | ||
713 | dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest); | |
714 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS); | |
715 | if (IS_ERR(req)) { | |
716 | d_drop(dentry); | |
717 | return PTR_ERR(req); | |
718 | } | |
719 | req->r_dentry = dget(dentry); | |
720 | req->r_num_caps = 2; | |
721 | req->r_path2 = kstrdup(dest, GFP_NOFS); | |
722 | req->r_locked_dir = dir; | |
723 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
724 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
725 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
726 | if (!err && !req->r_reply_info.head->is_dentry) | |
727 | err = ceph_handle_notrace_create(dir, dentry); | |
728 | ceph_mdsc_put_request(req); | |
729 | if (err) | |
730 | d_drop(dentry); | |
731 | return err; | |
732 | } | |
733 | ||
734 | static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
735 | { | |
3d14c5d2 YS |
736 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
737 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
738 | struct ceph_mds_request *req; |
739 | int err = -EROFS; | |
740 | int op; | |
741 | ||
742 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
743 | /* mkdir .snap/foo is a MKSNAP */ | |
744 | op = CEPH_MDS_OP_MKSNAP; | |
745 | dout("mksnap dir %p snap '%.*s' dn %p\n", dir, | |
746 | dentry->d_name.len, dentry->d_name.name, dentry); | |
747 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
748 | dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode); | |
749 | op = CEPH_MDS_OP_MKDIR; | |
750 | } else { | |
751 | goto out; | |
752 | } | |
753 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
754 | if (IS_ERR(req)) { | |
755 | err = PTR_ERR(req); | |
756 | goto out; | |
757 | } | |
758 | ||
759 | req->r_dentry = dget(dentry); | |
760 | req->r_num_caps = 2; | |
761 | req->r_locked_dir = dir; | |
762 | req->r_args.mkdir.mode = cpu_to_le32(mode); | |
763 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
764 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
765 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
766 | if (!err && !req->r_reply_info.head->is_dentry) | |
767 | err = ceph_handle_notrace_create(dir, dentry); | |
768 | ceph_mdsc_put_request(req); | |
769 | out: | |
770 | if (err < 0) | |
771 | d_drop(dentry); | |
772 | return err; | |
773 | } | |
774 | ||
775 | static int ceph_link(struct dentry *old_dentry, struct inode *dir, | |
776 | struct dentry *dentry) | |
777 | { | |
3d14c5d2 YS |
778 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
779 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
780 | struct ceph_mds_request *req; |
781 | int err; | |
782 | ||
783 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
784 | return -EROFS; | |
785 | ||
786 | dout("link in dir %p old_dentry %p dentry %p\n", dir, | |
787 | old_dentry, dentry); | |
788 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS); | |
789 | if (IS_ERR(req)) { | |
790 | d_drop(dentry); | |
791 | return PTR_ERR(req); | |
792 | } | |
793 | req->r_dentry = dget(dentry); | |
794 | req->r_num_caps = 2; | |
795 | req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */ | |
796 | req->r_locked_dir = dir; | |
797 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
798 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
799 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
70b666c3 | 800 | if (err) { |
2817b000 | 801 | d_drop(dentry); |
70b666c3 SW |
802 | } else if (!req->r_reply_info.head->is_dentry) { |
803 | ihold(old_dentry->d_inode); | |
804 | d_instantiate(dentry, old_dentry->d_inode); | |
805 | } | |
2817b000 SW |
806 | ceph_mdsc_put_request(req); |
807 | return err; | |
808 | } | |
809 | ||
810 | /* | |
811 | * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it | |
812 | * looks like the link count will hit 0, drop any other caps (other | |
813 | * than PIN) we don't specifically want (due to the file still being | |
814 | * open). | |
815 | */ | |
816 | static int drop_caps_for_unlink(struct inode *inode) | |
817 | { | |
818 | struct ceph_inode_info *ci = ceph_inode(inode); | |
819 | int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
820 | ||
821 | spin_lock(&inode->i_lock); | |
822 | if (inode->i_nlink == 1) { | |
823 | drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN); | |
824 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
825 | } | |
826 | spin_unlock(&inode->i_lock); | |
827 | return drop; | |
828 | } | |
829 | ||
830 | /* | |
831 | * rmdir and unlink are differ only by the metadata op code | |
832 | */ | |
833 | static int ceph_unlink(struct inode *dir, struct dentry *dentry) | |
834 | { | |
3d14c5d2 YS |
835 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
836 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
837 | struct inode *inode = dentry->d_inode; |
838 | struct ceph_mds_request *req; | |
839 | int err = -EROFS; | |
840 | int op; | |
841 | ||
842 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
843 | /* rmdir .snap/foo is RMSNAP */ | |
844 | dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len, | |
845 | dentry->d_name.name, dentry); | |
846 | op = CEPH_MDS_OP_RMSNAP; | |
847 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
848 | dout("unlink/rmdir dir %p dn %p inode %p\n", | |
849 | dir, dentry, inode); | |
850 | op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ? | |
851 | CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK; | |
852 | } else | |
853 | goto out; | |
854 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
855 | if (IS_ERR(req)) { | |
856 | err = PTR_ERR(req); | |
857 | goto out; | |
858 | } | |
859 | req->r_dentry = dget(dentry); | |
860 | req->r_num_caps = 2; | |
861 | req->r_locked_dir = dir; | |
862 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
863 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
864 | req->r_inode_drop = drop_caps_for_unlink(inode); | |
865 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
866 | if (!err && !req->r_reply_info.head->is_dentry) | |
867 | d_delete(dentry); | |
868 | ceph_mdsc_put_request(req); | |
869 | out: | |
870 | return err; | |
871 | } | |
872 | ||
873 | static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry, | |
874 | struct inode *new_dir, struct dentry *new_dentry) | |
875 | { | |
3d14c5d2 YS |
876 | struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb); |
877 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
878 | struct ceph_mds_request *req; |
879 | int err; | |
880 | ||
881 | if (ceph_snap(old_dir) != ceph_snap(new_dir)) | |
882 | return -EXDEV; | |
883 | if (ceph_snap(old_dir) != CEPH_NOSNAP || | |
884 | ceph_snap(new_dir) != CEPH_NOSNAP) | |
885 | return -EROFS; | |
886 | dout("rename dir %p dentry %p to dir %p dentry %p\n", | |
887 | old_dir, old_dentry, new_dir, new_dentry); | |
888 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS); | |
889 | if (IS_ERR(req)) | |
890 | return PTR_ERR(req); | |
891 | req->r_dentry = dget(new_dentry); | |
892 | req->r_num_caps = 2; | |
893 | req->r_old_dentry = dget(old_dentry); | |
894 | req->r_locked_dir = new_dir; | |
895 | req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
896 | req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
897 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
898 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
899 | /* release LINK_RDCACHE on source inode (mds will lock it) */ | |
900 | req->r_old_inode_drop = CEPH_CAP_LINK_SHARED; | |
901 | if (new_dentry->d_inode) | |
902 | req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode); | |
903 | err = ceph_mdsc_do_request(mdsc, old_dir, req); | |
904 | if (!err && !req->r_reply_info.head->is_dentry) { | |
905 | /* | |
906 | * Normally d_move() is done by fill_trace (called by | |
907 | * do_request, above). If there is no trace, we need | |
908 | * to do it here. | |
909 | */ | |
ea1409f9 SW |
910 | |
911 | /* d_move screws up d_subdirs order */ | |
912 | ceph_i_clear(new_dir, CEPH_I_COMPLETE); | |
913 | ||
2817b000 | 914 | d_move(old_dentry, new_dentry); |
ea1409f9 SW |
915 | |
916 | /* ensure target dentry is invalidated, despite | |
917 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d | 918 | ceph_invalidate_dentry_lease(new_dentry); |
2817b000 SW |
919 | } |
920 | ceph_mdsc_put_request(req); | |
921 | return err; | |
922 | } | |
923 | ||
81a6cf2d SW |
924 | /* |
925 | * Ensure a dentry lease will no longer revalidate. | |
926 | */ | |
927 | void ceph_invalidate_dentry_lease(struct dentry *dentry) | |
928 | { | |
929 | spin_lock(&dentry->d_lock); | |
930 | dentry->d_time = jiffies; | |
931 | ceph_dentry(dentry)->lease_shared_gen = 0; | |
932 | spin_unlock(&dentry->d_lock); | |
933 | } | |
2817b000 SW |
934 | |
935 | /* | |
936 | * Check if dentry lease is valid. If not, delete the lease. Try to | |
937 | * renew if the least is more than half up. | |
938 | */ | |
939 | static int dentry_lease_is_valid(struct dentry *dentry) | |
940 | { | |
941 | struct ceph_dentry_info *di; | |
942 | struct ceph_mds_session *s; | |
943 | int valid = 0; | |
944 | u32 gen; | |
945 | unsigned long ttl; | |
946 | struct ceph_mds_session *session = NULL; | |
947 | struct inode *dir = NULL; | |
948 | u32 seq = 0; | |
949 | ||
950 | spin_lock(&dentry->d_lock); | |
951 | di = ceph_dentry(dentry); | |
952 | if (di && di->lease_session) { | |
953 | s = di->lease_session; | |
954 | spin_lock(&s->s_cap_lock); | |
955 | gen = s->s_cap_gen; | |
956 | ttl = s->s_cap_ttl; | |
957 | spin_unlock(&s->s_cap_lock); | |
958 | ||
959 | if (di->lease_gen == gen && | |
960 | time_before(jiffies, dentry->d_time) && | |
961 | time_before(jiffies, ttl)) { | |
962 | valid = 1; | |
963 | if (di->lease_renew_after && | |
964 | time_after(jiffies, di->lease_renew_after)) { | |
965 | /* we should renew */ | |
966 | dir = dentry->d_parent->d_inode; | |
967 | session = ceph_get_mds_session(s); | |
968 | seq = di->lease_seq; | |
969 | di->lease_renew_after = 0; | |
970 | di->lease_renew_from = jiffies; | |
971 | } | |
2817b000 SW |
972 | } |
973 | } | |
974 | spin_unlock(&dentry->d_lock); | |
975 | ||
976 | if (session) { | |
977 | ceph_mdsc_lease_send_msg(session, dir, dentry, | |
978 | CEPH_MDS_LEASE_RENEW, seq); | |
979 | ceph_put_mds_session(session); | |
980 | } | |
981 | dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid); | |
982 | return valid; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Check if directory-wide content lease/cap is valid. | |
987 | */ | |
988 | static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry) | |
989 | { | |
990 | struct ceph_inode_info *ci = ceph_inode(dir); | |
991 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
992 | int valid = 0; | |
993 | ||
994 | spin_lock(&dir->i_lock); | |
995 | if (ci->i_shared_gen == di->lease_shared_gen) | |
996 | valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1); | |
997 | spin_unlock(&dir->i_lock); | |
998 | dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n", | |
999 | dir, (unsigned)ci->i_shared_gen, dentry, | |
1000 | (unsigned)di->lease_shared_gen, valid); | |
1001 | return valid; | |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * Check if cached dentry can be trusted. | |
1006 | */ | |
1007 | static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1008 | { | |
34286d66 NP |
1009 | struct inode *dir; |
1010 | ||
0eb980e3 | 1011 | if (nd && nd->flags & LOOKUP_RCU) |
34286d66 NP |
1012 | return -ECHILD; |
1013 | ||
1014 | dir = dentry->d_parent->d_inode; | |
2817b000 | 1015 | |
1cd3935b SW |
1016 | dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry, |
1017 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode, | |
1018 | ceph_dentry(dentry)->offset); | |
2817b000 SW |
1019 | |
1020 | /* always trust cached snapped dentries, snapdir dentry */ | |
1021 | if (ceph_snap(dir) != CEPH_NOSNAP) { | |
1022 | dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry, | |
1023 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode); | |
1024 | goto out_touch; | |
1025 | } | |
1026 | if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) | |
1027 | goto out_touch; | |
1028 | ||
1029 | if (dentry_lease_is_valid(dentry) || | |
1030 | dir_lease_is_valid(dir, dentry)) | |
1031 | goto out_touch; | |
1032 | ||
1033 | dout("d_revalidate %p invalid\n", dentry); | |
1034 | d_drop(dentry); | |
1035 | return 0; | |
1036 | out_touch: | |
1037 | ceph_dentry_lru_touch(dentry); | |
1038 | return 1; | |
1039 | } | |
1040 | ||
1041 | /* | |
147851d2 | 1042 | * Release our ceph_dentry_info. |
2817b000 | 1043 | */ |
147851d2 | 1044 | static void ceph_d_release(struct dentry *dentry) |
2817b000 SW |
1045 | { |
1046 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
2817b000 | 1047 | |
147851d2 | 1048 | dout("d_release %p\n", dentry); |
2817b000 SW |
1049 | if (di) { |
1050 | ceph_dentry_lru_del(dentry); | |
1051 | if (di->lease_session) | |
1052 | ceph_put_mds_session(di->lease_session); | |
1053 | kmem_cache_free(ceph_dentry_cachep, di); | |
1054 | dentry->d_fsdata = NULL; | |
1055 | } | |
1056 | } | |
1057 | ||
1058 | static int ceph_snapdir_d_revalidate(struct dentry *dentry, | |
1059 | struct nameidata *nd) | |
1060 | { | |
1061 | /* | |
1062 | * Eventually, we'll want to revalidate snapped metadata | |
1063 | * too... probably... | |
1064 | */ | |
1065 | return 1; | |
1066 | } | |
1067 | ||
1068 | ||
1069 | ||
1070 | /* | |
1071 | * read() on a dir. This weird interface hack only works if mounted | |
1072 | * with '-o dirstat'. | |
1073 | */ | |
1074 | static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size, | |
1075 | loff_t *ppos) | |
1076 | { | |
1077 | struct ceph_file_info *cf = file->private_data; | |
1078 | struct inode *inode = file->f_dentry->d_inode; | |
1079 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1080 | int left; | |
ae598083 | 1081 | const int bufsize = 1024; |
2817b000 | 1082 | |
3d14c5d2 | 1083 | if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT)) |
2817b000 SW |
1084 | return -EISDIR; |
1085 | ||
1086 | if (!cf->dir_info) { | |
ae598083 | 1087 | cf->dir_info = kmalloc(bufsize, GFP_NOFS); |
2817b000 SW |
1088 | if (!cf->dir_info) |
1089 | return -ENOMEM; | |
1090 | cf->dir_info_len = | |
ae598083 | 1091 | snprintf(cf->dir_info, bufsize, |
2817b000 SW |
1092 | "entries: %20lld\n" |
1093 | " files: %20lld\n" | |
1094 | " subdirs: %20lld\n" | |
1095 | "rentries: %20lld\n" | |
1096 | " rfiles: %20lld\n" | |
1097 | " rsubdirs: %20lld\n" | |
1098 | "rbytes: %20lld\n" | |
1099 | "rctime: %10ld.%09ld\n", | |
1100 | ci->i_files + ci->i_subdirs, | |
1101 | ci->i_files, | |
1102 | ci->i_subdirs, | |
1103 | ci->i_rfiles + ci->i_rsubdirs, | |
1104 | ci->i_rfiles, | |
1105 | ci->i_rsubdirs, | |
1106 | ci->i_rbytes, | |
1107 | (long)ci->i_rctime.tv_sec, | |
1108 | (long)ci->i_rctime.tv_nsec); | |
1109 | } | |
1110 | ||
1111 | if (*ppos >= cf->dir_info_len) | |
1112 | return 0; | |
1113 | size = min_t(unsigned, size, cf->dir_info_len-*ppos); | |
1114 | left = copy_to_user(buf, cf->dir_info + *ppos, size); | |
1115 | if (left == size) | |
1116 | return -EFAULT; | |
1117 | *ppos += (size - left); | |
1118 | return size - left; | |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * an fsync() on a dir will wait for any uncommitted directory | |
1123 | * operations to commit. | |
1124 | */ | |
7ea80859 | 1125 | static int ceph_dir_fsync(struct file *file, int datasync) |
2817b000 | 1126 | { |
7ea80859 | 1127 | struct inode *inode = file->f_path.dentry->d_inode; |
2817b000 SW |
1128 | struct ceph_inode_info *ci = ceph_inode(inode); |
1129 | struct list_head *head = &ci->i_unsafe_dirops; | |
1130 | struct ceph_mds_request *req; | |
1131 | u64 last_tid; | |
1132 | int ret = 0; | |
1133 | ||
1134 | dout("dir_fsync %p\n", inode); | |
1135 | spin_lock(&ci->i_unsafe_lock); | |
1136 | if (list_empty(head)) | |
1137 | goto out; | |
1138 | ||
1139 | req = list_entry(head->prev, | |
1140 | struct ceph_mds_request, r_unsafe_dir_item); | |
1141 | last_tid = req->r_tid; | |
1142 | ||
1143 | do { | |
1144 | ceph_mdsc_get_request(req); | |
1145 | spin_unlock(&ci->i_unsafe_lock); | |
1146 | dout("dir_fsync %p wait on tid %llu (until %llu)\n", | |
1147 | inode, req->r_tid, last_tid); | |
1148 | if (req->r_timeout) { | |
1149 | ret = wait_for_completion_timeout( | |
1150 | &req->r_safe_completion, req->r_timeout); | |
1151 | if (ret > 0) | |
1152 | ret = 0; | |
1153 | else if (ret == 0) | |
1154 | ret = -EIO; /* timed out */ | |
1155 | } else { | |
1156 | wait_for_completion(&req->r_safe_completion); | |
1157 | } | |
1158 | spin_lock(&ci->i_unsafe_lock); | |
1159 | ceph_mdsc_put_request(req); | |
1160 | ||
1161 | if (ret || list_empty(head)) | |
1162 | break; | |
1163 | req = list_entry(head->next, | |
1164 | struct ceph_mds_request, r_unsafe_dir_item); | |
1165 | } while (req->r_tid < last_tid); | |
1166 | out: | |
1167 | spin_unlock(&ci->i_unsafe_lock); | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * We maintain a private dentry LRU. | |
1173 | * | |
1174 | * FIXME: this needs to be changed to a per-mds lru to be useful. | |
1175 | */ | |
1176 | void ceph_dentry_lru_add(struct dentry *dn) | |
1177 | { | |
1178 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1179 | struct ceph_mds_client *mdsc; | |
2817b000 | 1180 | |
04a419f9 SW |
1181 | dout("dentry_lru_add %p %p '%.*s'\n", di, dn, |
1182 | dn->d_name.len, dn->d_name.name); | |
2817b000 | 1183 | if (di) { |
3d14c5d2 | 1184 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
2817b000 SW |
1185 | spin_lock(&mdsc->dentry_lru_lock); |
1186 | list_add_tail(&di->lru, &mdsc->dentry_lru); | |
1187 | mdsc->num_dentry++; | |
1188 | spin_unlock(&mdsc->dentry_lru_lock); | |
1189 | } | |
1190 | } | |
1191 | ||
1192 | void ceph_dentry_lru_touch(struct dentry *dn) | |
1193 | { | |
1194 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1195 | struct ceph_mds_client *mdsc; | |
2817b000 | 1196 | |
1cd3935b SW |
1197 | dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn, |
1198 | dn->d_name.len, dn->d_name.name, di->offset); | |
2817b000 | 1199 | if (di) { |
3d14c5d2 | 1200 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
2817b000 SW |
1201 | spin_lock(&mdsc->dentry_lru_lock); |
1202 | list_move_tail(&di->lru, &mdsc->dentry_lru); | |
1203 | spin_unlock(&mdsc->dentry_lru_lock); | |
1204 | } | |
1205 | } | |
1206 | ||
1207 | void ceph_dentry_lru_del(struct dentry *dn) | |
1208 | { | |
1209 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1210 | struct ceph_mds_client *mdsc; | |
1211 | ||
04a419f9 SW |
1212 | dout("dentry_lru_del %p %p '%.*s'\n", di, dn, |
1213 | dn->d_name.len, dn->d_name.name); | |
2817b000 | 1214 | if (di) { |
3d14c5d2 | 1215 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
2817b000 SW |
1216 | spin_lock(&mdsc->dentry_lru_lock); |
1217 | list_del_init(&di->lru); | |
1218 | mdsc->num_dentry--; | |
1219 | spin_unlock(&mdsc->dentry_lru_lock); | |
1220 | } | |
1221 | } | |
1222 | ||
6c0f3af7 SW |
1223 | /* |
1224 | * Return name hash for a given dentry. This is dependent on | |
1225 | * the parent directory's hash function. | |
1226 | */ | |
1227 | unsigned ceph_dentry_hash(struct dentry *dn) | |
1228 | { | |
1229 | struct inode *dir = dn->d_parent->d_inode; | |
1230 | struct ceph_inode_info *dci = ceph_inode(dir); | |
1231 | ||
1232 | switch (dci->i_dir_layout.dl_dir_hash) { | |
1233 | case 0: /* for backward compat */ | |
1234 | case CEPH_STR_HASH_LINUX: | |
1235 | return dn->d_name.hash; | |
1236 | ||
1237 | default: | |
1238 | return ceph_str_hash(dci->i_dir_layout.dl_dir_hash, | |
1239 | dn->d_name.name, dn->d_name.len); | |
1240 | } | |
1241 | } | |
1242 | ||
2817b000 SW |
1243 | const struct file_operations ceph_dir_fops = { |
1244 | .read = ceph_read_dir, | |
1245 | .readdir = ceph_readdir, | |
1246 | .llseek = ceph_dir_llseek, | |
1247 | .open = ceph_open, | |
1248 | .release = ceph_release, | |
1249 | .unlocked_ioctl = ceph_ioctl, | |
1250 | .fsync = ceph_dir_fsync, | |
1251 | }; | |
1252 | ||
1253 | const struct inode_operations ceph_dir_iops = { | |
1254 | .lookup = ceph_lookup, | |
1255 | .permission = ceph_permission, | |
1256 | .getattr = ceph_getattr, | |
1257 | .setattr = ceph_setattr, | |
1258 | .setxattr = ceph_setxattr, | |
1259 | .getxattr = ceph_getxattr, | |
1260 | .listxattr = ceph_listxattr, | |
1261 | .removexattr = ceph_removexattr, | |
1262 | .mknod = ceph_mknod, | |
1263 | .symlink = ceph_symlink, | |
1264 | .mkdir = ceph_mkdir, | |
1265 | .link = ceph_link, | |
1266 | .unlink = ceph_unlink, | |
1267 | .rmdir = ceph_unlink, | |
1268 | .rename = ceph_rename, | |
1269 | .create = ceph_create, | |
1270 | }; | |
1271 | ||
52dfb8ac | 1272 | const struct dentry_operations ceph_dentry_ops = { |
2817b000 | 1273 | .d_revalidate = ceph_d_revalidate, |
147851d2 | 1274 | .d_release = ceph_d_release, |
2817b000 SW |
1275 | }; |
1276 | ||
52dfb8ac | 1277 | const struct dentry_operations ceph_snapdir_dentry_ops = { |
2817b000 | 1278 | .d_revalidate = ceph_snapdir_d_revalidate, |
147851d2 | 1279 | .d_release = ceph_d_release, |
2817b000 SW |
1280 | }; |
1281 | ||
52dfb8ac | 1282 | const struct dentry_operations ceph_snap_dentry_ops = { |
147851d2 | 1283 | .d_release = ceph_d_release, |
2817b000 | 1284 | }; |