blob: fd11fb231a2ea796e86eae933265a488ad6774e5 [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
Sage Weil52dfb8a2010-08-03 10:25:30 -070029const struct dentry_operations ceph_dentry_ops;
Sage Weil2817b002009-10-06 11:31:08 -070030
31/*
32 * Initialize ceph dentry state.
33 */
34int ceph_init_dentry(struct dentry *dentry)
35{
36 struct ceph_dentry_info *di;
37
38 if (dentry->d_fsdata)
39 return 0;
40
Yan, Zheng687265e2015-06-13 17:27:05 +080041 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_KERNEL | __GFP_ZERO);
Sage Weil2817b002009-10-06 11:31:08 -070042 if (!di)
43 return -ENOMEM; /* oh well */
44
45 spin_lock(&dentry->d_lock);
Sage Weil8c6efb52010-04-23 11:36:54 -070046 if (dentry->d_fsdata) {
47 /* lost a race */
48 kmem_cache_free(ceph_dentry_cachep, di);
Sage Weil2817b002009-10-06 11:31:08 -070049 goto out_unlock;
Sage Weil8c6efb52010-04-23 11:36:54 -070050 }
Sage Weil48d0cbd2011-07-26 11:30:15 -070051
David Howells2b0143b2015-03-17 22:25:59 +000052 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP)
Sage Weil48d0cbd2011-07-26 11:30:15 -070053 d_set_d_op(dentry, &ceph_dentry_ops);
David Howells2b0143b2015-03-17 22:25:59 +000054 else if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
Sage Weil48d0cbd2011-07-26 11:30:15 -070055 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
56 else
57 d_set_d_op(dentry, &ceph_snap_dentry_ops);
58
Sage Weil2817b002009-10-06 11:31:08 -070059 di->dentry = dentry;
60 di->lease_session = NULL;
Sage Weil2817b002009-10-06 11:31:08 -070061 dentry->d_time = jiffies;
Sage Weil48d0cbd2011-07-26 11:30:15 -070062 /* avoid reordering d_fsdata setup so that the check above is safe */
63 smp_mb();
64 dentry->d_fsdata = di;
Sage Weil2817b002009-10-06 11:31:08 -070065 ceph_dentry_lru_add(dentry);
66out_unlock:
67 spin_unlock(&dentry->d_lock);
68 return 0;
69}
70
Sage Weil5f21c962011-07-26 11:30:29 -070071struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
72{
73 struct inode *inode = NULL;
74
75 if (!dentry)
76 return NULL;
77
78 spin_lock(&dentry->d_lock);
Sage Weil8842b3b2012-06-07 13:43:35 -070079 if (!IS_ROOT(dentry)) {
David Howells2b0143b2015-03-17 22:25:59 +000080 inode = d_inode(dentry->d_parent);
Sage Weil5f21c962011-07-26 11:30:29 -070081 ihold(inode);
82 }
83 spin_unlock(&dentry->d_lock);
84 return inode;
85}
Sage Weil2817b002009-10-06 11:31:08 -070086
87
88/*
89 * for readdir, we encode the directory frag and offset within that
90 * frag into f_pos.
91 */
92static unsigned fpos_frag(loff_t p)
93{
94 return p >> 32;
95}
96static unsigned fpos_off(loff_t p)
97{
98 return p & 0xffffffff;
99}
100
Yan, Zheng4d5f5df2014-02-13 19:40:26 +0800101static int fpos_cmp(loff_t l, loff_t r)
102{
103 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
104 if (v)
105 return v;
106 return (int)(fpos_off(l) - fpos_off(r));
107}
108
Sage Weil2817b002009-10-06 11:31:08 -0700109/*
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800110 * make note of the last dentry we read, so we can
111 * continue at the same lexicographical point,
112 * regardless of what dir changes take place on the
113 * server.
114 */
115static int note_last_dentry(struct ceph_file_info *fi, const char *name,
116 int len, unsigned next_offset)
117{
118 char *buf = kmalloc(len+1, GFP_KERNEL);
119 if (!buf)
120 return -ENOMEM;
121 kfree(fi->last_name);
122 fi->last_name = buf;
123 memcpy(fi->last_name, name, len);
124 fi->last_name[len] = 0;
125 fi->next_offset = next_offset;
126 dout("note_last_dentry '%s'\n", fi->last_name);
127 return 0;
128}
129
130/*
Sage Weil2817b002009-10-06 11:31:08 -0700131 * When possible, we try to satisfy a readdir by peeking at the
132 * dcache. We make this work by carefully ordering dentries on
Al Viro946e51f2014-10-26 19:19:16 -0400133 * d_child when we initially get results back from the MDS, and
Sage Weil2817b002009-10-06 11:31:08 -0700134 * falling back to a "normal" sync readdir if any dentries in the dir
135 * are dropped.
136 *
Yan, Zheng2f276c52013-03-13 19:44:32 +0800137 * Complete dir indicates that we have all dentries in the dir. It is
Sage Weil2817b002009-10-06 11:31:08 -0700138 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
139 * the MDS if/when the directory is modified).
140 */
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800141static int __dcache_readdir(struct file *file, struct dir_context *ctx,
142 u32 shared_gen)
Sage Weil2817b002009-10-06 11:31:08 -0700143{
Al Viro77acfa22013-05-17 16:52:26 -0400144 struct ceph_file_info *fi = file->private_data;
Al Virob5830432014-10-31 01:22:04 -0400145 struct dentry *parent = file->f_path.dentry;
David Howells2b0143b2015-03-17 22:25:59 +0000146 struct inode *dir = d_inode(parent);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800147 struct dentry *dentry, *last = NULL;
Sage Weil2817b002009-10-06 11:31:08 -0700148 struct ceph_dentry_info *di;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800149 unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry *);
Sage Weil2817b002009-10-06 11:31:08 -0700150 int err = 0;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800151 loff_t ptr_pos = 0;
152 struct ceph_readdir_cache_control cache_ctl = {};
Sage Weil2817b002009-10-06 11:31:08 -0700153
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800154 dout("__dcache_readdir %p v%u at %llu\n", dir, shared_gen, ctx->pos);
Sage Weil2817b002009-10-06 11:31:08 -0700155
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800156 /* we can calculate cache index for the first dirfrag */
157 if (ceph_frag_is_leftmost(fpos_frag(ctx->pos))) {
158 cache_ctl.index = fpos_off(ctx->pos) - 2;
159 BUG_ON(cache_ctl.index < 0);
160 ptr_pos = cache_ctl.index * sizeof(struct dentry *);
Sage Weil2817b002009-10-06 11:31:08 -0700161 }
162
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800163 while (true) {
164 pgoff_t pgoff;
165 bool emit_dentry;
166
167 if (ptr_pos >= i_size_read(dir)) {
Sage Weil9cfa1092011-07-26 11:26:18 -0700168 fi->flags |= CEPH_F_ATEND;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800169 err = 0;
170 break;
Sage Weil2817b002009-10-06 11:31:08 -0700171 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800172
173 err = -EAGAIN;
174 pgoff = ptr_pos >> PAGE_CACHE_SHIFT;
175 if (!cache_ctl.page || pgoff != page_index(cache_ctl.page)) {
176 ceph_readdir_cache_release(&cache_ctl);
177 cache_ctl.page = find_lock_page(&dir->i_data, pgoff);
178 if (!cache_ctl.page) {
179 dout(" page %lu not found\n", pgoff);
180 break;
181 }
182 /* reading/filling the cache are serialized by
183 * i_mutex, no need to use page lock */
184 unlock_page(cache_ctl.page);
185 cache_ctl.dentries = kmap(cache_ctl.page);
186 }
187
188 rcu_read_lock();
189 spin_lock(&parent->d_lock);
190 /* check i_size again here, because empty directory can be
191 * marked as complete while not holding the i_mutex. */
192 if (ceph_dir_is_complete_ordered(dir) &&
193 ptr_pos < i_size_read(dir))
194 dentry = cache_ctl.dentries[cache_ctl.index % nsize];
195 else
196 dentry = NULL;
197 spin_unlock(&parent->d_lock);
198 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
199 dentry = NULL;
200 rcu_read_unlock();
201 if (!dentry)
202 break;
203
204 emit_dentry = false;
205 di = ceph_dentry(dentry);
206 spin_lock(&dentry->d_lock);
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800207 if (di->lease_shared_gen == shared_gen &&
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800208 d_really_is_positive(dentry) &&
David Howells2b0143b2015-03-17 22:25:59 +0000209 ceph_snap(d_inode(dentry)) != CEPH_SNAPDIR &&
210 ceph_ino(d_inode(dentry)) != CEPH_INO_CEPH &&
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800211 fpos_cmp(ctx->pos, di->offset) <= 0) {
212 emit_dentry = true;
Sage Weil2817b002009-10-06 11:31:08 -0700213 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800214 spin_unlock(&dentry->d_lock);
215
216 if (emit_dentry) {
217 dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
218 dentry, dentry, d_inode(dentry));
219 ctx->pos = di->offset;
220 if (!dir_emit(ctx, dentry->d_name.name,
221 dentry->d_name.len,
222 ceph_translate_ino(dentry->d_sb,
223 d_inode(dentry)->i_ino),
224 d_inode(dentry)->i_mode >> 12)) {
225 dput(dentry);
226 err = 0;
227 break;
228 }
229 ctx->pos++;
230
231 if (last)
232 dput(last);
233 last = dentry;
234 } else {
235 dput(dentry);
236 }
237
238 cache_ctl.index++;
239 ptr_pos += sizeof(struct dentry *);
Sage Weil2817b002009-10-06 11:31:08 -0700240 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800241 ceph_readdir_cache_release(&cache_ctl);
242 if (last) {
243 int ret;
244 di = ceph_dentry(last);
245 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
246 fpos_off(di->offset) + 1);
247 if (ret < 0)
248 err = ret;
Al Viro77acfa22013-05-17 16:52:26 -0400249 dput(last);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800250 }
Sage Weil2817b002009-10-06 11:31:08 -0700251 return err;
252}
253
Al Viro77acfa22013-05-17 16:52:26 -0400254static int ceph_readdir(struct file *file, struct dir_context *ctx)
Sage Weil2817b002009-10-06 11:31:08 -0700255{
Al Viro77acfa22013-05-17 16:52:26 -0400256 struct ceph_file_info *fi = file->private_data;
257 struct inode *inode = file_inode(file);
Sage Weil2817b002009-10-06 11:31:08 -0700258 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700259 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
260 struct ceph_mds_client *mdsc = fsc->mdsc;
Al Viro77acfa22013-05-17 16:52:26 -0400261 unsigned frag = fpos_frag(ctx->pos);
262 int off = fpos_off(ctx->pos);
Sage Weil2817b002009-10-06 11:31:08 -0700263 int err;
264 u32 ftype;
265 struct ceph_mds_reply_info_parsed *rinfo;
Sage Weil2817b002009-10-06 11:31:08 -0700266
Al Viro77acfa22013-05-17 16:52:26 -0400267 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
Sage Weil9cfa1092011-07-26 11:26:18 -0700268 if (fi->flags & CEPH_F_ATEND)
Sage Weil2817b002009-10-06 11:31:08 -0700269 return 0;
270
271 /* always start with . and .. */
Al Viro77acfa22013-05-17 16:52:26 -0400272 if (ctx->pos == 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700273 dout("readdir off 0 -> '.'\n");
Al Viro77acfa22013-05-17 16:52:26 -0400274 if (!dir_emit(ctx, ".", 1,
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800275 ceph_translate_ino(inode->i_sb, inode->i_ino),
Al Viro77acfa22013-05-17 16:52:26 -0400276 inode->i_mode >> 12))
Sage Weil2817b002009-10-06 11:31:08 -0700277 return 0;
Al Viro77acfa22013-05-17 16:52:26 -0400278 ctx->pos = 1;
Sage Weil2817b002009-10-06 11:31:08 -0700279 off = 1;
280 }
Al Viro77acfa22013-05-17 16:52:26 -0400281 if (ctx->pos == 1) {
Al Virob5830432014-10-31 01:22:04 -0400282 ino_t ino = parent_ino(file->f_path.dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700283 dout("readdir off 1 -> '..'\n");
Al Viro77acfa22013-05-17 16:52:26 -0400284 if (!dir_emit(ctx, "..", 2,
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800285 ceph_translate_ino(inode->i_sb, ino),
Al Viro77acfa22013-05-17 16:52:26 -0400286 inode->i_mode >> 12))
Sage Weil2817b002009-10-06 11:31:08 -0700287 return 0;
Al Viro77acfa22013-05-17 16:52:26 -0400288 ctx->pos = 2;
Sage Weil2817b002009-10-06 11:31:08 -0700289 off = 2;
290 }
291
292 /* can we use the dcache? */
Sage Weilbe655592011-11-30 09:47:09 -0800293 spin_lock(&ci->i_ceph_lock);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800294 if (ceph_test_mount_opt(fsc, DCACHE) &&
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700295 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
Sage Weila0dff782010-07-22 13:47:21 -0700296 ceph_snap(inode) != CEPH_SNAPDIR &&
Yan, Zheng70db4f32014-10-21 18:09:56 -0700297 __ceph_dir_is_complete_ordered(ci) &&
Sage Weil2817b002009-10-06 11:31:08 -0700298 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800299 u32 shared_gen = ci->i_shared_gen;
Sage Weilbe655592011-11-30 09:47:09 -0800300 spin_unlock(&ci->i_ceph_lock);
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800301 err = __dcache_readdir(file, ctx, shared_gen);
Sage Weilefa4c122010-10-18 14:04:31 -0700302 if (err != -EAGAIN)
Sage Weil2817b002009-10-06 11:31:08 -0700303 return err;
Yan, Zheng0081bd82014-04-08 21:42:59 +0800304 frag = fpos_frag(ctx->pos);
305 off = fpos_off(ctx->pos);
Sage Weilefa4c122010-10-18 14:04:31 -0700306 } else {
Sage Weilbe655592011-11-30 09:47:09 -0800307 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700308 }
Sage Weil2817b002009-10-06 11:31:08 -0700309
310 /* proceed with a normal readdir */
Sage Weil2817b002009-10-06 11:31:08 -0700311more:
312 /* do we have the correct frag content buffered? */
313 if (fi->frag != frag || fi->last_readdir == NULL) {
314 struct ceph_mds_request *req;
315 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
316 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
317
318 /* discard old result, if any */
Sage Weil393f6622010-03-10 12:03:32 -0800319 if (fi->last_readdir) {
Sage Weil2817b002009-10-06 11:31:08 -0700320 ceph_mdsc_put_request(fi->last_readdir);
Sage Weil393f6622010-03-10 12:03:32 -0800321 fi->last_readdir = NULL;
322 }
Sage Weil2817b002009-10-06 11:31:08 -0700323
Sage Weil2817b002009-10-06 11:31:08 -0700324 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);
Yan, Zheng54008392014-03-29 13:41:15 +0800329 err = ceph_alloc_readdir_reply_buffer(req, inode);
330 if (err) {
331 ceph_mdsc_put_request(req);
332 return err;
333 }
Sage Weil2817b002009-10-06 11:31:08 -0700334 /* hints to request -> mds selection code */
335 req->r_direct_mode = USE_AUTH_MDS;
336 req->r_direct_hash = ceph_frag_value(frag);
337 req->r_direct_is_hash = true;
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400338 if (fi->last_name) {
Yan, Zheng687265e2015-06-13 17:27:05 +0800339 req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400340 if (!req->r_path2) {
341 ceph_mdsc_put_request(req);
342 return -ENOMEM;
343 }
344 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800345 req->r_dir_release_cnt = fi->dir_release_count;
346 req->r_dir_ordered_cnt = fi->dir_ordered_count;
347 req->r_readdir_cache_idx = fi->readdir_cache_idx;
Sage Weil2817b002009-10-06 11:31:08 -0700348 req->r_readdir_offset = fi->next_offset;
349 req->r_args.readdir.frag = cpu_to_le32(frag);
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400350
351 req->r_inode = inode;
352 ihold(inode);
353 req->r_dentry = dget(file->f_path.dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700354 err = ceph_mdsc_do_request(mdsc, NULL, req);
355 if (err < 0) {
356 ceph_mdsc_put_request(req);
357 return err;
358 }
359 dout("readdir got and parsed readdir result=%d"
360 " on frag %x, end=%d, complete=%d\n", err, frag,
361 (int)req->r_reply_info.dir_end,
362 (int)req->r_reply_info.dir_complete);
363
Sage Weil2817b002009-10-06 11:31:08 -0700364
365 /* note next offset and last dentry name */
Yan, Zheng81c6aea2013-09-18 09:44:13 +0800366 rinfo = &req->r_reply_info;
367 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
368 frag = le32_to_cpu(rinfo->dir_dir->frag);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800369 off = req->r_readdir_offset;
370 fi->next_offset = off;
Yan, Zheng81c6aea2013-09-18 09:44:13 +0800371 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800372
Yan, Zhengf0494202014-02-27 16:26:24 +0800373 fi->frag = frag;
Sage Weil2817b002009-10-06 11:31:08 -0700374 fi->offset = fi->next_offset;
375 fi->last_readdir = req;
376
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800377 if (req->r_did_prepopulate) {
378 fi->readdir_cache_idx = req->r_readdir_cache_idx;
379 if (fi->readdir_cache_idx < 0) {
380 /* preclude from marking dir ordered */
381 fi->dir_ordered_count = 0;
382 } else if (ceph_frag_is_leftmost(frag) && off == 2) {
383 /* note dir version at start of readdir so
384 * we can tell if any dentries get dropped */
385 fi->dir_release_count = req->r_dir_release_cnt;
386 fi->dir_ordered_count = req->r_dir_ordered_cnt;
387 }
388 } else {
389 dout("readdir !did_prepopulate");
390 /* disable readdir cache */
391 fi->readdir_cache_idx = -1;
392 /* preclude from marking dir complete */
393 fi->dir_release_count = 0;
394 }
395
Sage Weil2817b002009-10-06 11:31:08 -0700396 if (req->r_reply_info.dir_end) {
397 kfree(fi->last_name);
398 fi->last_name = NULL;
Sage Weil7b88dad2010-11-11 16:48:59 -0800399 if (ceph_frag_is_rightmost(frag))
400 fi->next_offset = 2;
401 else
402 fi->next_offset = 0;
Sage Weil2817b002009-10-06 11:31:08 -0700403 } else {
Sage Weil2817b002009-10-06 11:31:08 -0700404 err = note_last_dentry(fi,
405 rinfo->dir_dname[rinfo->dir_nr-1],
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800406 rinfo->dir_dname_len[rinfo->dir_nr-1],
407 fi->next_offset + rinfo->dir_nr);
Sage Weil2817b002009-10-06 11:31:08 -0700408 if (err)
409 return err;
Sage Weil2817b002009-10-06 11:31:08 -0700410 }
411 }
412
413 rinfo = &fi->last_readdir->r_reply_info;
414 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
415 rinfo->dir_nr, off, fi->offset);
Al Viro77acfa22013-05-17 16:52:26 -0400416
417 ctx->pos = ceph_make_fpos(frag, off);
Sage Weilda398222011-05-12 15:28:11 -0700418 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
Sage Weil2817b002009-10-06 11:31:08 -0700419 struct ceph_mds_reply_inode *in =
420 rinfo->dir_in[off - fi->offset].in;
Sage Weil3105c192010-11-18 09:15:07 -0800421 struct ceph_vino vino;
422 ino_t ino;
423
Sage Weil2817b002009-10-06 11:31:08 -0700424 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
Al Viro77acfa22013-05-17 16:52:26 -0400425 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
Sage Weil2817b002009-10-06 11:31:08 -0700426 rinfo->dir_dname_len[off - fi->offset],
427 rinfo->dir_dname[off - fi->offset], in);
428 BUG_ON(!in);
429 ftype = le32_to_cpu(in->mode) >> 12;
Sage Weil3105c192010-11-18 09:15:07 -0800430 vino.ino = le64_to_cpu(in->ino);
431 vino.snap = le64_to_cpu(in->snapid);
432 ino = ceph_vino_to_ino(vino);
Al Viro77acfa22013-05-17 16:52:26 -0400433 if (!dir_emit(ctx,
Sage Weil2817b002009-10-06 11:31:08 -0700434 rinfo->dir_dname[off - fi->offset],
435 rinfo->dir_dname_len[off - fi->offset],
Al Viro77acfa22013-05-17 16:52:26 -0400436 ceph_translate_ino(inode->i_sb, ino), ftype)) {
Sage Weil2817b002009-10-06 11:31:08 -0700437 dout("filldir stopping us...\n");
438 return 0;
439 }
440 off++;
Al Viro77acfa22013-05-17 16:52:26 -0400441 ctx->pos++;
Sage Weil2817b002009-10-06 11:31:08 -0700442 }
443
444 if (fi->last_name) {
445 ceph_mdsc_put_request(fi->last_readdir);
446 fi->last_readdir = NULL;
447 goto more;
448 }
449
450 /* more frags? */
451 if (!ceph_frag_is_rightmost(frag)) {
452 frag = ceph_frag_next(frag);
453 off = 0;
Al Viro77acfa22013-05-17 16:52:26 -0400454 ctx->pos = ceph_make_fpos(frag, off);
Sage Weil2817b002009-10-06 11:31:08 -0700455 dout("readdir next frag is %x\n", frag);
456 goto more;
457 }
Sage Weil9cfa1092011-07-26 11:26:18 -0700458 fi->flags |= CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700459
460 /*
461 * if dir_release_count still matches the dir, no dentries
462 * were released during the whole readdir, and we should have
463 * the complete dir contents in our cache.
464 */
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800465 if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
466 spin_lock(&ci->i_ceph_lock);
467 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
Yan, Zheng70db4f32014-10-21 18:09:56 -0700468 dout(" marking %p complete and ordered\n", inode);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800469 /* use i_size to track number of entries in
470 * readdir cache */
471 BUG_ON(fi->readdir_cache_idx < 0);
472 i_size_write(inode, fi->readdir_cache_idx *
473 sizeof(struct dentry*));
474 } else {
Yan, Zheng70db4f32014-10-21 18:09:56 -0700475 dout(" marking %p complete\n", inode);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800476 }
Yan, Zheng70db4f32014-10-21 18:09:56 -0700477 __ceph_dir_set_complete(ci, fi->dir_release_count,
478 fi->dir_ordered_count);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800479 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700480 }
Sage Weil2817b002009-10-06 11:31:08 -0700481
Al Viro77acfa22013-05-17 16:52:26 -0400482 dout("readdir %p file %p done.\n", inode, file);
Sage Weil2817b002009-10-06 11:31:08 -0700483 return 0;
484}
485
Yan, Zhengdcd3cc02014-02-28 16:36:09 +0800486static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
Sage Weil2817b002009-10-06 11:31:08 -0700487{
488 if (fi->last_readdir) {
489 ceph_mdsc_put_request(fi->last_readdir);
490 fi->last_readdir = NULL;
491 }
492 kfree(fi->last_name);
Sage Weila1629c32010-11-11 15:24:06 -0800493 fi->last_name = NULL;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800494 fi->dir_release_count = 0;
495 fi->readdir_cache_idx = -1;
Yan, Zhengdcd3cc02014-02-28 16:36:09 +0800496 if (ceph_frag_is_leftmost(frag))
497 fi->next_offset = 2; /* compensate for . and .. */
498 else
499 fi->next_offset = 0;
Sage Weil9cfa1092011-07-26 11:26:18 -0700500 fi->flags &= ~CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700501}
502
Andrew Morton965c8e52012-12-17 15:59:39 -0800503static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
Sage Weil2817b002009-10-06 11:31:08 -0700504{
505 struct ceph_file_info *fi = file->private_data;
506 struct inode *inode = file->f_mapping->host;
Yan, Zhengf0494202014-02-27 16:26:24 +0800507 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
Sage Weil2817b002009-10-06 11:31:08 -0700508 loff_t retval;
509
Al Viro59551022016-01-22 15:40:57 -0500510 inode_lock(inode);
Josef Bacik06222e42011-07-18 13:21:38 -0400511 retval = -EINVAL;
Andrew Morton965c8e52012-12-17 15:59:39 -0800512 switch (whence) {
Sage Weil2817b002009-10-06 11:31:08 -0700513 case SEEK_CUR:
514 offset += file->f_pos;
Josef Bacik06222e42011-07-18 13:21:38 -0400515 case SEEK_SET:
516 break;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800517 case SEEK_END:
518 retval = -EOPNOTSUPP;
Josef Bacik06222e42011-07-18 13:21:38 -0400519 default:
520 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700521 }
Josef Bacik06222e42011-07-18 13:21:38 -0400522
Yan, Zhengf0494202014-02-27 16:26:24 +0800523 if (offset >= 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700524 if (offset != file->f_pos) {
525 file->f_pos = offset;
526 file->f_version = 0;
Sage Weil9cfa1092011-07-26 11:26:18 -0700527 fi->flags &= ~CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700528 }
529 retval = offset;
530
Sage Weil2817b002009-10-06 11:31:08 -0700531 if (offset == 0 ||
Yan, Zhengf0494202014-02-27 16:26:24 +0800532 fpos_frag(offset) != fi->frag ||
Sage Weil2817b002009-10-06 11:31:08 -0700533 fpos_off(offset) < fi->offset) {
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800534 /* discard buffered readdir content on seekdir(0), or
535 * seek to new frag, or seek prior to current chunk */
Sage Weil2817b002009-10-06 11:31:08 -0700536 dout("dir_llseek dropping %p content\n", file);
Yan, Zhengdcd3cc02014-02-28 16:36:09 +0800537 reset_readdir(fi, fpos_frag(offset));
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800538 } else if (fpos_cmp(offset, old_offset) > 0) {
539 /* reset dir_release_count if we did a forward seek */
540 fi->dir_release_count = 0;
541 fi->readdir_cache_idx = -1;
Sage Weil2817b002009-10-06 11:31:08 -0700542 }
Sage Weil2817b002009-10-06 11:31:08 -0700543 }
Josef Bacik06222e42011-07-18 13:21:38 -0400544out:
Al Viro59551022016-01-22 15:40:57 -0500545 inode_unlock(inode);
Sage Weil2817b002009-10-06 11:31:08 -0700546 return retval;
547}
548
549/*
Sage Weil468640e2011-07-26 11:28:11 -0700550 * Handle lookups for the hidden .snap directory.
Sage Weil2817b002009-10-06 11:31:08 -0700551 */
Sage Weil468640e2011-07-26 11:28:11 -0700552int ceph_handle_snapdir(struct ceph_mds_request *req,
553 struct dentry *dentry, int err)
Sage Weil2817b002009-10-06 11:31:08 -0700554{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700555 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
David Howells2b0143b2015-03-17 22:25:59 +0000556 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
Sage Weil2817b002009-10-06 11:31:08 -0700557
558 /* .snap dir? */
559 if (err == -ENOENT &&
Sage Weil455cec02011-03-03 13:44:35 -0800560 ceph_snap(parent) == CEPH_NOSNAP &&
Sage Weil6b805182009-10-27 11:50:50 -0700561 strcmp(dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700562 fsc->mount_options->snapdir_name) == 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700563 struct inode *inode = ceph_get_snapdir(parent);
Al Viroa4555892014-10-21 20:11:25 -0400564 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
565 dentry, dentry, inode);
Sage Weil9358c6d2010-03-30 13:54:41 -0700566 BUG_ON(!d_unhashed(dentry));
Sage Weil2817b002009-10-06 11:31:08 -0700567 d_add(dentry, inode);
568 err = 0;
569 }
Sage Weil468640e2011-07-26 11:28:11 -0700570 return err;
571}
Sage Weil2817b002009-10-06 11:31:08 -0700572
Sage Weil468640e2011-07-26 11:28:11 -0700573/*
574 * Figure out final result of a lookup/open request.
575 *
576 * Mainly, make sure we return the final req->r_dentry (if it already
577 * existed) in place of the original VFS-provided dentry when they
578 * differ.
579 *
580 * Gracefully handle the case where the MDS replies with -ENOENT and
581 * no trace (which it may do, at its discretion, e.g., if it doesn't
582 * care to issue a lease on the negative dentry).
583 */
584struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
585 struct dentry *dentry, int err)
586{
Sage Weil2817b002009-10-06 11:31:08 -0700587 if (err == -ENOENT) {
588 /* no trace? */
589 err = 0;
590 if (!req->r_reply_info.head->is_dentry) {
591 dout("ENOENT and no trace, dentry %p inode %p\n",
David Howells2b0143b2015-03-17 22:25:59 +0000592 dentry, d_inode(dentry));
593 if (d_really_is_positive(dentry)) {
Sage Weil2817b002009-10-06 11:31:08 -0700594 d_drop(dentry);
595 err = -ENOENT;
596 } else {
597 d_add(dentry, NULL);
598 }
599 }
600 }
601 if (err)
602 dentry = ERR_PTR(err);
603 else if (dentry != req->r_dentry)
604 dentry = dget(req->r_dentry); /* we got spliced */
605 else
606 dentry = NULL;
607 return dentry;
608}
609
Sage Weil1d1de9162009-12-02 11:54:25 -0800610static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
611{
612 return ceph_ino(inode) == CEPH_INO_ROOT &&
613 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
614}
615
Sage Weil2817b002009-10-06 11:31:08 -0700616/*
617 * Look up a single dir entry. If there is a lookup intent, inform
618 * the MDS so that it gets our 'caps wanted' value in a single op.
619 */
620static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
Al Viro00cd8dd2012-06-10 17:13:09 -0400621 unsigned int flags)
Sage Weil2817b002009-10-06 11:31:08 -0700622{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700623 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
624 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700625 struct ceph_mds_request *req;
626 int op;
627 int err;
628
Al Viroa4555892014-10-21 20:11:25 -0400629 dout("lookup %p dentry %p '%pd'\n",
630 dir, dentry, dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700631
632 if (dentry->d_name.len > NAME_MAX)
633 return ERR_PTR(-ENAMETOOLONG);
634
635 err = ceph_init_dentry(dentry);
636 if (err < 0)
637 return ERR_PTR(err);
638
Sage Weil2817b002009-10-06 11:31:08 -0700639 /* can we conclude ENOENT locally? */
David Howells2b0143b2015-03-17 22:25:59 +0000640 if (d_really_is_negative(dentry)) {
Sage Weil2817b002009-10-06 11:31:08 -0700641 struct ceph_inode_info *ci = ceph_inode(dir);
642 struct ceph_dentry_info *di = ceph_dentry(dentry);
643
Sage Weilbe655592011-11-30 09:47:09 -0800644 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700645 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
646 if (strncmp(dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700647 fsc->mount_options->snapdir_name,
Sage Weil2817b002009-10-06 11:31:08 -0700648 dentry->d_name.len) &&
Sage Weil1d1de9162009-12-02 11:54:25 -0800649 !is_root_ceph_dentry(dir, dentry) &&
Yan, Zhenge2c3de02015-03-04 16:05:04 +0800650 ceph_test_mount_opt(fsc, DCACHE) &&
Yan, Zheng2f276c52013-03-13 19:44:32 +0800651 __ceph_dir_is_complete(ci) &&
Sage Weil2817b002009-10-06 11:31:08 -0700652 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
Sage Weilbe655592011-11-30 09:47:09 -0800653 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700654 dout(" dir %p complete, -ENOENT\n", dir);
655 d_add(dentry, NULL);
656 di->lease_shared_gen = ci->i_shared_gen;
657 return NULL;
658 }
Sage Weilbe655592011-11-30 09:47:09 -0800659 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700660 }
661
662 op = ceph_snap(dir) == CEPH_SNAPDIR ?
663 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
664 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
665 if (IS_ERR(req))
Julia Lawall7e34bc52010-05-22 12:01:14 +0200666 return ERR_CAST(req);
Sage Weil2817b002009-10-06 11:31:08 -0700667 req->r_dentry = dget(dentry);
668 req->r_num_caps = 2;
669 /* we only need inode linkage */
670 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
671 req->r_locked_dir = dir;
672 err = ceph_mdsc_do_request(mdsc, NULL, req);
Sage Weil468640e2011-07-26 11:28:11 -0700673 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil2817b002009-10-06 11:31:08 -0700674 dentry = ceph_finish_lookup(req, dentry, err);
675 ceph_mdsc_put_request(req); /* will dput(dentry) */
676 dout("lookup result=%p\n", dentry);
677 return dentry;
678}
679
680/*
681 * If we do a create but get no trace back from the MDS, follow up with
682 * a lookup (the VFS expects us to link up the provided dentry).
683 */
684int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
685{
Al Viro00cd8dd2012-06-10 17:13:09 -0400686 struct dentry *result = ceph_lookup(dir, dentry, 0);
Sage Weil2817b002009-10-06 11:31:08 -0700687
688 if (result && !IS_ERR(result)) {
689 /*
690 * We created the item, then did a lookup, and found
691 * it was already linked to another inode we already
Yan, Zheng4d41cef2015-02-04 15:10:48 +0800692 * had in our cache (and thus got spliced). To not
693 * confuse VFS (especially when inode is a directory),
694 * we don't link our dentry to that inode, return an
695 * error instead.
696 *
697 * This event should be rare and it happens only when
698 * we talk to old MDS. Recent MDS does not send traceless
699 * reply for request that creates new inode.
Sage Weil2817b002009-10-06 11:31:08 -0700700 */
Yan, Zheng5cba3722015-02-02 11:27:56 +0800701 d_drop(result);
Yan, Zheng4d41cef2015-02-04 15:10:48 +0800702 return -ESTALE;
Sage Weil2817b002009-10-06 11:31:08 -0700703 }
704 return PTR_ERR(result);
705}
706
707static int ceph_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -0400708 umode_t mode, dev_t rdev)
Sage Weil2817b002009-10-06 11:31:08 -0700709{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700710 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
711 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700712 struct ceph_mds_request *req;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800713 struct ceph_acls_info acls = {};
Sage Weil2817b002009-10-06 11:31:08 -0700714 int err;
715
716 if (ceph_snap(dir) != CEPH_NOSNAP)
717 return -EROFS;
718
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800719 err = ceph_pre_init_acls(dir, &mode, &acls);
720 if (err < 0)
721 return err;
722
Al Viro1a67aaf2011-07-26 01:52:52 -0400723 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
Sage Weil2817b002009-10-06 11:31:08 -0700724 dir, dentry, mode, rdev);
725 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
726 if (IS_ERR(req)) {
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800727 err = PTR_ERR(req);
728 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700729 }
730 req->r_dentry = dget(dentry);
731 req->r_num_caps = 2;
732 req->r_locked_dir = dir;
733 req->r_args.mknod.mode = cpu_to_le32(mode);
734 req->r_args.mknod.rdev = cpu_to_le32(rdev);
735 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
736 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800737 if (acls.pagelist) {
738 req->r_pagelist = acls.pagelist;
739 acls.pagelist = NULL;
740 }
Sage Weil2817b002009-10-06 11:31:08 -0700741 err = ceph_mdsc_do_request(mdsc, dir, req);
742 if (!err && !req->r_reply_info.head->is_dentry)
743 err = ceph_handle_notrace_create(dir, dentry);
744 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800745out:
Guangliang Zhao7221fe42013-11-11 15:18:03 +0800746 if (!err)
David Howells2b0143b2015-03-17 22:25:59 +0000747 ceph_init_inode_acls(d_inode(dentry), &acls);
Yan, Zhengb20a95a2014-02-11 12:55:05 +0800748 else
Sage Weil2817b002009-10-06 11:31:08 -0700749 d_drop(dentry);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800750 ceph_release_acls_info(&acls);
Sage Weil2817b002009-10-06 11:31:08 -0700751 return err;
752}
753
Al Viro4acdaf22011-07-26 01:42:34 -0400754static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -0400755 bool excl)
Sage Weil2817b002009-10-06 11:31:08 -0700756{
Miklos Szeredi2d83bde2012-06-05 15:10:25 +0200757 return ceph_mknod(dir, dentry, mode, 0);
Sage Weil2817b002009-10-06 11:31:08 -0700758}
759
760static int ceph_symlink(struct inode *dir, struct dentry *dentry,
761 const char *dest)
762{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700763 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
764 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700765 struct ceph_mds_request *req;
766 int err;
767
768 if (ceph_snap(dir) != CEPH_NOSNAP)
769 return -EROFS;
770
771 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
772 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
773 if (IS_ERR(req)) {
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800774 err = PTR_ERR(req);
775 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700776 }
Yan, Zheng687265e2015-06-13 17:27:05 +0800777 req->r_path2 = kstrdup(dest, GFP_KERNEL);
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400778 if (!req->r_path2) {
779 err = -ENOMEM;
780 ceph_mdsc_put_request(req);
781 goto out;
782 }
783 req->r_locked_dir = dir;
Sage Weil2817b002009-10-06 11:31:08 -0700784 req->r_dentry = dget(dentry);
785 req->r_num_caps = 2;
Sage Weil2817b002009-10-06 11:31:08 -0700786 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
787 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
788 err = ceph_mdsc_do_request(mdsc, dir, req);
789 if (!err && !req->r_reply_info.head->is_dentry)
790 err = ceph_handle_notrace_create(dir, dentry);
791 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800792out:
793 if (err)
Sage Weil2817b002009-10-06 11:31:08 -0700794 d_drop(dentry);
795 return err;
796}
797
Al Viro18bb1db2011-07-26 01:41:39 -0400798static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Sage Weil2817b002009-10-06 11:31:08 -0700799{
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;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800803 struct ceph_acls_info acls = {};
Sage Weil2817b002009-10-06 11:31:08 -0700804 int err = -EROFS;
805 int op;
806
807 if (ceph_snap(dir) == CEPH_SNAPDIR) {
808 /* mkdir .snap/foo is a MKSNAP */
809 op = CEPH_MDS_OP_MKSNAP;
Al Viroa4555892014-10-21 20:11:25 -0400810 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
811 dentry, dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700812 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
Al Viro18bb1db2011-07-26 01:41:39 -0400813 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
Sage Weil2817b002009-10-06 11:31:08 -0700814 op = CEPH_MDS_OP_MKDIR;
815 } else {
816 goto out;
817 }
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800818
819 mode |= S_IFDIR;
820 err = ceph_pre_init_acls(dir, &mode, &acls);
821 if (err < 0)
822 goto out;
823
Sage Weil2817b002009-10-06 11:31:08 -0700824 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
825 if (IS_ERR(req)) {
826 err = PTR_ERR(req);
827 goto out;
828 }
829
830 req->r_dentry = dget(dentry);
831 req->r_num_caps = 2;
832 req->r_locked_dir = dir;
833 req->r_args.mkdir.mode = cpu_to_le32(mode);
834 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
835 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800836 if (acls.pagelist) {
837 req->r_pagelist = acls.pagelist;
838 acls.pagelist = NULL;
839 }
Sage Weil2817b002009-10-06 11:31:08 -0700840 err = ceph_mdsc_do_request(mdsc, dir, req);
Yan, Zheng275dd192014-12-10 16:17:31 +0800841 if (!err &&
842 !req->r_reply_info.head->is_target &&
843 !req->r_reply_info.head->is_dentry)
Sage Weil2817b002009-10-06 11:31:08 -0700844 err = ceph_handle_notrace_create(dir, dentry);
845 ceph_mdsc_put_request(req);
846out:
Yan, Zhengb20a95a2014-02-11 12:55:05 +0800847 if (!err)
David Howells2b0143b2015-03-17 22:25:59 +0000848 ceph_init_inode_acls(d_inode(dentry), &acls);
Yan, Zhengb20a95a2014-02-11 12:55:05 +0800849 else
Sage Weil2817b002009-10-06 11:31:08 -0700850 d_drop(dentry);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800851 ceph_release_acls_info(&acls);
Sage Weil2817b002009-10-06 11:31:08 -0700852 return err;
853}
854
855static int ceph_link(struct dentry *old_dentry, struct inode *dir,
856 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 ceph_mds_request *req;
861 int err;
862
863 if (ceph_snap(dir) != CEPH_NOSNAP)
864 return -EROFS;
865
866 dout("link in dir %p old_dentry %p dentry %p\n", dir,
867 old_dentry, dentry);
868 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
869 if (IS_ERR(req)) {
870 d_drop(dentry);
871 return PTR_ERR(req);
872 }
873 req->r_dentry = dget(dentry);
874 req->r_num_caps = 2;
Sage Weil4b58c9b2013-02-05 13:41:23 -0800875 req->r_old_dentry = dget(old_dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700876 req->r_locked_dir = dir;
877 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
878 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengad88f232013-07-21 20:25:25 +0800879 /* release LINK_SHARED on source inode (mds will lock it) */
880 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
Sage Weil2817b002009-10-06 11:31:08 -0700881 err = ceph_mdsc_do_request(mdsc, dir, req);
Sage Weil70b666c2011-05-27 09:24:26 -0700882 if (err) {
Sage Weil2817b002009-10-06 11:31:08 -0700883 d_drop(dentry);
Sage Weil70b666c2011-05-27 09:24:26 -0700884 } else if (!req->r_reply_info.head->is_dentry) {
David Howells2b0143b2015-03-17 22:25:59 +0000885 ihold(d_inode(old_dentry));
886 d_instantiate(dentry, d_inode(old_dentry));
Sage Weil70b666c2011-05-27 09:24:26 -0700887 }
Sage Weil2817b002009-10-06 11:31:08 -0700888 ceph_mdsc_put_request(req);
889 return err;
890}
891
892/*
893 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
894 * looks like the link count will hit 0, drop any other caps (other
895 * than PIN) we don't specifically want (due to the file still being
896 * open).
897 */
898static int drop_caps_for_unlink(struct inode *inode)
899{
900 struct ceph_inode_info *ci = ceph_inode(inode);
901 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
902
Sage Weilbe655592011-11-30 09:47:09 -0800903 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700904 if (inode->i_nlink == 1) {
905 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
906 ci->i_ceph_flags |= CEPH_I_NODELAY;
907 }
Sage Weilbe655592011-11-30 09:47:09 -0800908 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700909 return drop;
910}
911
912/*
913 * rmdir and unlink are differ only by the metadata op code
914 */
915static int ceph_unlink(struct inode *dir, struct dentry *dentry)
916{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700917 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
918 struct ceph_mds_client *mdsc = fsc->mdsc;
David Howells2b0143b2015-03-17 22:25:59 +0000919 struct inode *inode = d_inode(dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700920 struct ceph_mds_request *req;
921 int err = -EROFS;
922 int op;
923
924 if (ceph_snap(dir) == CEPH_SNAPDIR) {
925 /* rmdir .snap/foo is RMSNAP */
Al Viroa4555892014-10-21 20:11:25 -0400926 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700927 op = CEPH_MDS_OP_RMSNAP;
928 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
929 dout("unlink/rmdir dir %p dn %p inode %p\n",
930 dir, dentry, inode);
David Howellse36cb0b2015-01-29 12:02:35 +0000931 op = d_is_dir(dentry) ?
Sage Weil2817b002009-10-06 11:31:08 -0700932 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
933 } else
934 goto out;
935 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
936 if (IS_ERR(req)) {
937 err = PTR_ERR(req);
938 goto out;
939 }
940 req->r_dentry = dget(dentry);
941 req->r_num_caps = 2;
942 req->r_locked_dir = dir;
943 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
944 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
945 req->r_inode_drop = drop_caps_for_unlink(inode);
946 err = ceph_mdsc_do_request(mdsc, dir, req);
947 if (!err && !req->r_reply_info.head->is_dentry)
948 d_delete(dentry);
949 ceph_mdsc_put_request(req);
950out:
951 return err;
952}
953
954static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
955 struct inode *new_dir, struct dentry *new_dentry)
956{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700957 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
958 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700959 struct ceph_mds_request *req;
Yan, Zheng0ea611a2015-04-07 15:36:32 +0800960 int op = CEPH_MDS_OP_RENAME;
Sage Weil2817b002009-10-06 11:31:08 -0700961 int err;
962
963 if (ceph_snap(old_dir) != ceph_snap(new_dir))
964 return -EXDEV;
Yan, Zheng0ea611a2015-04-07 15:36:32 +0800965 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
966 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
967 op = CEPH_MDS_OP_RENAMESNAP;
968 else
969 return -EROFS;
970 }
Sage Weil2817b002009-10-06 11:31:08 -0700971 dout("rename dir %p dentry %p to dir %p dentry %p\n",
972 old_dir, old_dentry, new_dir, new_dentry);
Yan, Zheng0ea611a2015-04-07 15:36:32 +0800973 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
Sage Weil2817b002009-10-06 11:31:08 -0700974 if (IS_ERR(req))
975 return PTR_ERR(req);
Sage Weil180061a2013-02-05 13:36:05 -0800976 ihold(old_dir);
Sage Weil2817b002009-10-06 11:31:08 -0700977 req->r_dentry = dget(new_dentry);
978 req->r_num_caps = 2;
979 req->r_old_dentry = dget(old_dentry);
Sage Weil180061a2013-02-05 13:36:05 -0800980 req->r_old_dentry_dir = old_dir;
Sage Weil2817b002009-10-06 11:31:08 -0700981 req->r_locked_dir = new_dir;
982 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
983 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
984 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
985 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
986 /* release LINK_RDCACHE on source inode (mds will lock it) */
987 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
David Howells2b0143b2015-03-17 22:25:59 +0000988 if (d_really_is_positive(new_dentry))
989 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
Sage Weil2817b002009-10-06 11:31:08 -0700990 err = ceph_mdsc_do_request(mdsc, old_dir, req);
991 if (!err && !req->r_reply_info.head->is_dentry) {
992 /*
993 * Normally d_move() is done by fill_trace (called by
994 * do_request, above). If there is no trace, we need
995 * to do it here.
996 */
Sage Weilea1409f2010-04-28 16:12:06 -0700997
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800998 /* d_move screws up sibling dentries' offsets */
999 ceph_dir_clear_complete(old_dir);
1000 ceph_dir_clear_complete(new_dir);
1001
Sage Weil2817b002009-10-06 11:31:08 -07001002 d_move(old_dentry, new_dentry);
Sage Weilea1409f2010-04-28 16:12:06 -07001003
1004 /* ensure target dentry is invalidated, despite
1005 rehashing bug in vfs_rename_dir */
Sage Weil81a6cf22010-05-14 09:35:38 -07001006 ceph_invalidate_dentry_lease(new_dentry);
Sage Weil2817b002009-10-06 11:31:08 -07001007 }
1008 ceph_mdsc_put_request(req);
1009 return err;
1010}
1011
Sage Weil81a6cf22010-05-14 09:35:38 -07001012/*
1013 * Ensure a dentry lease will no longer revalidate.
1014 */
1015void ceph_invalidate_dentry_lease(struct dentry *dentry)
1016{
1017 spin_lock(&dentry->d_lock);
1018 dentry->d_time = jiffies;
1019 ceph_dentry(dentry)->lease_shared_gen = 0;
1020 spin_unlock(&dentry->d_lock);
1021}
Sage Weil2817b002009-10-06 11:31:08 -07001022
1023/*
1024 * Check if dentry lease is valid. If not, delete the lease. Try to
1025 * renew if the least is more than half up.
1026 */
1027static int dentry_lease_is_valid(struct dentry *dentry)
1028{
1029 struct ceph_dentry_info *di;
1030 struct ceph_mds_session *s;
1031 int valid = 0;
1032 u32 gen;
1033 unsigned long ttl;
1034 struct ceph_mds_session *session = NULL;
1035 struct inode *dir = NULL;
1036 u32 seq = 0;
1037
1038 spin_lock(&dentry->d_lock);
1039 di = ceph_dentry(dentry);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001040 if (di->lease_session) {
Sage Weil2817b002009-10-06 11:31:08 -07001041 s = di->lease_session;
Alex Elderd8fb02a2012-01-12 17:48:10 -08001042 spin_lock(&s->s_gen_ttl_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001043 gen = s->s_cap_gen;
1044 ttl = s->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -08001045 spin_unlock(&s->s_gen_ttl_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001046
1047 if (di->lease_gen == gen &&
1048 time_before(jiffies, dentry->d_time) &&
1049 time_before(jiffies, ttl)) {
1050 valid = 1;
1051 if (di->lease_renew_after &&
1052 time_after(jiffies, di->lease_renew_after)) {
1053 /* we should renew */
David Howells2b0143b2015-03-17 22:25:59 +00001054 dir = d_inode(dentry->d_parent);
Sage Weil2817b002009-10-06 11:31:08 -07001055 session = ceph_get_mds_session(s);
1056 seq = di->lease_seq;
1057 di->lease_renew_after = 0;
1058 di->lease_renew_from = jiffies;
1059 }
Sage Weil2817b002009-10-06 11:31:08 -07001060 }
1061 }
1062 spin_unlock(&dentry->d_lock);
1063
1064 if (session) {
1065 ceph_mdsc_lease_send_msg(session, dir, dentry,
1066 CEPH_MDS_LEASE_RENEW, seq);
1067 ceph_put_mds_session(session);
1068 }
1069 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1070 return valid;
1071}
1072
1073/*
1074 * Check if directory-wide content lease/cap is valid.
1075 */
1076static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1077{
1078 struct ceph_inode_info *ci = ceph_inode(dir);
1079 struct ceph_dentry_info *di = ceph_dentry(dentry);
1080 int valid = 0;
1081
Sage Weilbe655592011-11-30 09:47:09 -08001082 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001083 if (ci->i_shared_gen == di->lease_shared_gen)
1084 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
Sage Weilbe655592011-11-30 09:47:09 -08001085 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001086 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1087 dir, (unsigned)ci->i_shared_gen, dentry,
1088 (unsigned)di->lease_shared_gen, valid);
1089 return valid;
1090}
1091
1092/*
1093 * Check if cached dentry can be trusted.
1094 */
Al Viro0b728e12012-06-10 16:03:43 -04001095static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
Sage Weil2817b002009-10-06 11:31:08 -07001096{
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001097 int valid = 0;
Nick Piggin34286d62011-01-07 17:49:57 +11001098 struct inode *dir;
1099
Al Viro0b728e12012-06-10 16:03:43 -04001100 if (flags & LOOKUP_RCU)
Nick Piggin34286d62011-01-07 17:49:57 +11001101 return -ECHILD;
1102
Al Viroa4555892014-10-21 20:11:25 -04001103 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
David Howells2b0143b2015-03-17 22:25:59 +00001104 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
Sage Weil2817b002009-10-06 11:31:08 -07001105
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001106 dir = ceph_get_dentry_parent_inode(dentry);
1107
Sage Weil2817b002009-10-06 11:31:08 -07001108 /* always trust cached snapped dentries, snapdir dentry */
1109 if (ceph_snap(dir) != CEPH_NOSNAP) {
Al Viroa4555892014-10-21 20:11:25 -04001110 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
David Howells2b0143b2015-03-17 22:25:59 +00001111 dentry, d_inode(dentry));
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001112 valid = 1;
David Howells2b0143b2015-03-17 22:25:59 +00001113 } else if (d_really_is_positive(dentry) &&
1114 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001115 valid = 1;
1116 } else if (dentry_lease_is_valid(dentry) ||
1117 dir_lease_is_valid(dir, dentry)) {
David Howells2b0143b2015-03-17 22:25:59 +00001118 if (d_really_is_positive(dentry))
1119 valid = ceph_is_any_caps(d_inode(dentry));
Yan, Zheng9215aee2013-11-30 12:47:41 +08001120 else
1121 valid = 1;
Sage Weil2817b002009-10-06 11:31:08 -07001122 }
Sage Weil2817b002009-10-06 11:31:08 -07001123
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001124 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
Yan, Zheng9215aee2013-11-30 12:47:41 +08001125 if (valid) {
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001126 ceph_dentry_lru_touch(dentry);
Yan, Zheng9215aee2013-11-30 12:47:41 +08001127 } else {
1128 ceph_dir_clear_complete(dir);
Yan, Zheng9215aee2013-11-30 12:47:41 +08001129 }
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001130 iput(dir);
1131 return valid;
Sage Weil2817b002009-10-06 11:31:08 -07001132}
1133
1134/*
Sage Weil147851d2011-03-15 14:57:41 -07001135 * Release our ceph_dentry_info.
Sage Weil2817b002009-10-06 11:31:08 -07001136 */
Sage Weil147851d2011-03-15 14:57:41 -07001137static void ceph_d_release(struct dentry *dentry)
Sage Weil2817b002009-10-06 11:31:08 -07001138{
1139 struct ceph_dentry_info *di = ceph_dentry(dentry);
Sage Weil2817b002009-10-06 11:31:08 -07001140
Sage Weil147851d2011-03-15 14:57:41 -07001141 dout("d_release %p\n", dentry);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001142 ceph_dentry_lru_del(dentry);
1143 if (di->lease_session)
1144 ceph_put_mds_session(di->lease_session);
1145 kmem_cache_free(ceph_dentry_cachep, di);
1146 dentry->d_fsdata = NULL;
Sage Weil2817b002009-10-06 11:31:08 -07001147}
1148
1149static int ceph_snapdir_d_revalidate(struct dentry *dentry,
Al Viro0b728e12012-06-10 16:03:43 -04001150 unsigned int flags)
Sage Weil2817b002009-10-06 11:31:08 -07001151{
1152 /*
1153 * Eventually, we'll want to revalidate snapped metadata
1154 * too... probably...
1155 */
1156 return 1;
1157}
1158
Sage Weilb58dc412011-03-15 15:53:40 -07001159/*
1160 * When the VFS prunes a dentry from the cache, we need to clear the
1161 * complete flag on the parent directory.
1162 *
1163 * Called under dentry->d_lock.
1164 */
1165static void ceph_d_prune(struct dentry *dentry)
1166{
Sage Weil774ac212011-11-11 09:48:08 -08001167 dout("ceph_d_prune %p\n", dentry);
Sage Weilb58dc412011-03-15 15:53:40 -07001168
1169 /* do we have a valid parent? */
Sage Weil8842b3b2012-06-07 13:43:35 -07001170 if (IS_ROOT(dentry))
Sage Weilb58dc412011-03-15 15:53:40 -07001171 return;
1172
Yan, Zheng2f276c52013-03-13 19:44:32 +08001173 /* if we are not hashed, we don't affect dir's completeness */
Sage Weilb58dc412011-03-15 15:53:40 -07001174 if (d_unhashed(dentry))
1175 return;
1176
1177 /*
1178 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1179 * cleared until d_release
1180 */
David Howells2b0143b2015-03-17 22:25:59 +00001181 ceph_dir_clear_complete(d_inode(dentry->d_parent));
Sage Weilb58dc412011-03-15 15:53:40 -07001182}
Sage Weil2817b002009-10-06 11:31:08 -07001183
1184/*
1185 * read() on a dir. This weird interface hack only works if mounted
1186 * with '-o dirstat'.
1187 */
1188static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1189 loff_t *ppos)
1190{
1191 struct ceph_file_info *cf = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -05001192 struct inode *inode = file_inode(file);
Sage Weil2817b002009-10-06 11:31:08 -07001193 struct ceph_inode_info *ci = ceph_inode(inode);
1194 int left;
Sage Weilae598082011-05-12 14:28:05 -07001195 const int bufsize = 1024;
Sage Weil2817b002009-10-06 11:31:08 -07001196
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001197 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
Sage Weil2817b002009-10-06 11:31:08 -07001198 return -EISDIR;
1199
1200 if (!cf->dir_info) {
Yan, Zheng687265e2015-06-13 17:27:05 +08001201 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
Sage Weil2817b002009-10-06 11:31:08 -07001202 if (!cf->dir_info)
1203 return -ENOMEM;
1204 cf->dir_info_len =
Sage Weilae598082011-05-12 14:28:05 -07001205 snprintf(cf->dir_info, bufsize,
Sage Weil2817b002009-10-06 11:31:08 -07001206 "entries: %20lld\n"
1207 " files: %20lld\n"
1208 " subdirs: %20lld\n"
1209 "rentries: %20lld\n"
1210 " rfiles: %20lld\n"
1211 " rsubdirs: %20lld\n"
1212 "rbytes: %20lld\n"
1213 "rctime: %10ld.%09ld\n",
1214 ci->i_files + ci->i_subdirs,
1215 ci->i_files,
1216 ci->i_subdirs,
1217 ci->i_rfiles + ci->i_rsubdirs,
1218 ci->i_rfiles,
1219 ci->i_rsubdirs,
1220 ci->i_rbytes,
1221 (long)ci->i_rctime.tv_sec,
1222 (long)ci->i_rctime.tv_nsec);
1223 }
1224
1225 if (*ppos >= cf->dir_info_len)
1226 return 0;
1227 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1228 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1229 if (left == size)
1230 return -EFAULT;
1231 *ppos += (size - left);
1232 return size - left;
1233}
1234
1235/*
Sage Weil2817b002009-10-06 11:31:08 -07001236 * We maintain a private dentry LRU.
1237 *
1238 * FIXME: this needs to be changed to a per-mds lru to be useful.
1239 */
1240void ceph_dentry_lru_add(struct dentry *dn)
1241{
1242 struct ceph_dentry_info *di = ceph_dentry(dn);
1243 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001244
Al Viroa4555892014-10-21 20:11:25 -04001245 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001246 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1247 spin_lock(&mdsc->dentry_lru_lock);
1248 list_add_tail(&di->lru, &mdsc->dentry_lru);
1249 mdsc->num_dentry++;
1250 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001251}
1252
1253void ceph_dentry_lru_touch(struct dentry *dn)
1254{
1255 struct ceph_dentry_info *di = ceph_dentry(dn);
1256 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001257
Al Viroa4555892014-10-21 20:11:25 -04001258 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1259 di->offset);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001260 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1261 spin_lock(&mdsc->dentry_lru_lock);
1262 list_move_tail(&di->lru, &mdsc->dentry_lru);
1263 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001264}
1265
1266void ceph_dentry_lru_del(struct dentry *dn)
1267{
1268 struct ceph_dentry_info *di = ceph_dentry(dn);
1269 struct ceph_mds_client *mdsc;
1270
Al Viroa4555892014-10-21 20:11:25 -04001271 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001272 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1273 spin_lock(&mdsc->dentry_lru_lock);
1274 list_del_init(&di->lru);
1275 mdsc->num_dentry--;
1276 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001277}
1278
Sage Weil6c0f3af2010-11-16 11:14:34 -08001279/*
1280 * Return name hash for a given dentry. This is dependent on
1281 * the parent directory's hash function.
1282 */
Sage Weile5f86dc2011-07-26 11:30:55 -07001283unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
Sage Weil6c0f3af2010-11-16 11:14:34 -08001284{
Sage Weil6c0f3af2010-11-16 11:14:34 -08001285 struct ceph_inode_info *dci = ceph_inode(dir);
1286
1287 switch (dci->i_dir_layout.dl_dir_hash) {
1288 case 0: /* for backward compat */
1289 case CEPH_STR_HASH_LINUX:
1290 return dn->d_name.hash;
1291
1292 default:
1293 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1294 dn->d_name.name, dn->d_name.len);
1295 }
1296}
1297
Sage Weil2817b002009-10-06 11:31:08 -07001298const struct file_operations ceph_dir_fops = {
1299 .read = ceph_read_dir,
Al Viro77acfa22013-05-17 16:52:26 -04001300 .iterate = ceph_readdir,
Sage Weil2817b002009-10-06 11:31:08 -07001301 .llseek = ceph_dir_llseek,
1302 .open = ceph_open,
1303 .release = ceph_release,
1304 .unlocked_ioctl = ceph_ioctl,
Yan, Zhengda819c82015-05-27 11:19:34 +08001305 .fsync = ceph_fsync,
Sage Weil2817b002009-10-06 11:31:08 -07001306};
1307
Yan, Zheng38c48b52015-01-14 13:46:04 +08001308const struct file_operations ceph_snapdir_fops = {
1309 .iterate = ceph_readdir,
1310 .llseek = ceph_dir_llseek,
1311 .open = ceph_open,
1312 .release = ceph_release,
1313};
1314
Sage Weil2817b002009-10-06 11:31:08 -07001315const struct inode_operations ceph_dir_iops = {
1316 .lookup = ceph_lookup,
1317 .permission = ceph_permission,
1318 .getattr = ceph_getattr,
1319 .setattr = ceph_setattr,
1320 .setxattr = ceph_setxattr,
1321 .getxattr = ceph_getxattr,
1322 .listxattr = ceph_listxattr,
1323 .removexattr = ceph_removexattr,
Guangliang Zhao7221fe42013-11-11 15:18:03 +08001324 .get_acl = ceph_get_acl,
Sage Weil72466d02014-01-29 06:22:25 -08001325 .set_acl = ceph_set_acl,
Sage Weil2817b002009-10-06 11:31:08 -07001326 .mknod = ceph_mknod,
1327 .symlink = ceph_symlink,
1328 .mkdir = ceph_mkdir,
1329 .link = ceph_link,
1330 .unlink = ceph_unlink,
1331 .rmdir = ceph_unlink,
1332 .rename = ceph_rename,
1333 .create = ceph_create,
Miklos Szeredi2d83bde2012-06-05 15:10:25 +02001334 .atomic_open = ceph_atomic_open,
Sage Weil2817b002009-10-06 11:31:08 -07001335};
1336
Yan, Zheng38c48b52015-01-14 13:46:04 +08001337const struct inode_operations ceph_snapdir_iops = {
1338 .lookup = ceph_lookup,
1339 .permission = ceph_permission,
1340 .getattr = ceph_getattr,
1341 .mkdir = ceph_mkdir,
1342 .rmdir = ceph_unlink,
Yan, Zheng0ea611a2015-04-07 15:36:32 +08001343 .rename = ceph_rename,
Yan, Zheng38c48b52015-01-14 13:46:04 +08001344};
1345
Sage Weil52dfb8a2010-08-03 10:25:30 -07001346const struct dentry_operations ceph_dentry_ops = {
Sage Weil2817b002009-10-06 11:31:08 -07001347 .d_revalidate = ceph_d_revalidate,
Sage Weil147851d2011-03-15 14:57:41 -07001348 .d_release = ceph_d_release,
Sage Weilb58dc412011-03-15 15:53:40 -07001349 .d_prune = ceph_d_prune,
Sage Weil2817b002009-10-06 11:31:08 -07001350};
1351
Sage Weil52dfb8a2010-08-03 10:25:30 -07001352const struct dentry_operations ceph_snapdir_dentry_ops = {
Sage Weil2817b002009-10-06 11:31:08 -07001353 .d_revalidate = ceph_snapdir_d_revalidate,
Sage Weil147851d2011-03-15 14:57:41 -07001354 .d_release = ceph_d_release,
Sage Weil2817b002009-10-06 11:31:08 -07001355};
1356
Sage Weil52dfb8a2010-08-03 10:25:30 -07001357const struct dentry_operations ceph_snap_dentry_ops = {
Sage Weil147851d2011-03-15 14:57:41 -07001358 .d_release = ceph_d_release,
Sage Weilb58dc412011-03-15 15:53:40 -07001359 .d_prune = ceph_d_prune,
Sage Weil2817b002009-10-06 11:31:08 -07001360};