Commit | Line | Data |
---|---|---|
5db11c21 MM |
1 | /* |
2 | * (C) 2001 Clemson University and The University of Chicago | |
3 | * | |
4 | * See COPYING in top-level directory. | |
5 | */ | |
6 | ||
7 | #include "protocol.h" | |
8 | #include "pvfs2-kernel.h" | |
9 | #include "pvfs2-bufmap.h" | |
10 | ||
11 | struct readdir_handle_s { | |
12 | int buffer_index; | |
13 | struct pvfs2_readdir_response_s readdir_response; | |
14 | void *dents_buf; | |
15 | }; | |
16 | ||
17 | /* | |
18 | * decode routine needed by kmod to make sense of the shared page for readdirs. | |
19 | */ | |
20 | static long decode_dirents(char *ptr, struct pvfs2_readdir_response_s *readdir) | |
21 | { | |
22 | int i; | |
23 | struct pvfs2_readdir_response_s *rd = | |
24 | (struct pvfs2_readdir_response_s *) ptr; | |
25 | char *buf = ptr; | |
5db11c21 MM |
26 | |
27 | readdir->token = rd->token; | |
28 | readdir->pvfs_dirent_outcount = rd->pvfs_dirent_outcount; | |
ef4af94e | 29 | readdir->dirent_array = kcalloc(readdir->pvfs_dirent_outcount, |
5db11c21 MM |
30 | sizeof(*readdir->dirent_array), |
31 | GFP_KERNEL); | |
32 | if (readdir->dirent_array == NULL) | |
33 | return -ENOMEM; | |
9be68b08 | 34 | buf += offsetof(struct pvfs2_readdir_response_s, dirent_array); |
5db11c21 | 35 | for (i = 0; i < readdir->pvfs_dirent_outcount; i++) { |
9be68b08 AV |
36 | __u32 len = *(__u32 *)buf; |
37 | readdir->dirent_array[i].d_name = buf + 4; | |
38 | buf += roundup8(4 + len + 1); | |
39 | readdir->dirent_array[i].d_length = len; | |
5db11c21 | 40 | readdir->dirent_array[i].khandle = |
9be68b08 AV |
41 | *(struct pvfs2_khandle *) buf; |
42 | buf += 16; | |
5db11c21 | 43 | } |
9be68b08 | 44 | return buf - ptr; |
5db11c21 MM |
45 | } |
46 | ||
47 | static long readdir_handle_ctor(struct readdir_handle_s *rhandle, void *buf, | |
48 | int buffer_index) | |
49 | { | |
50 | long ret; | |
51 | ||
52 | if (buf == NULL) { | |
53 | gossip_err | |
54 | ("Invalid NULL buffer specified in readdir_handle_ctor\n"); | |
55 | return -ENOMEM; | |
56 | } | |
57 | if (buffer_index < 0) { | |
58 | gossip_err | |
59 | ("Invalid buffer index specified in readdir_handle_ctor\n"); | |
60 | return -EINVAL; | |
61 | } | |
62 | rhandle->buffer_index = buffer_index; | |
63 | rhandle->dents_buf = buf; | |
64 | ret = decode_dirents(buf, &rhandle->readdir_response); | |
65 | if (ret < 0) { | |
66 | gossip_err("Could not decode readdir from buffer %ld\n", ret); | |
67 | rhandle->buffer_index = -1; | |
68 | gossip_debug(GOSSIP_DIR_DEBUG, "vfree %p\n", buf); | |
69 | vfree(buf); | |
70 | rhandle->dents_buf = NULL; | |
71 | } | |
72 | return ret; | |
73 | } | |
74 | ||
75 | static void readdir_handle_dtor(struct pvfs2_bufmap *bufmap, | |
76 | struct readdir_handle_s *rhandle) | |
77 | { | |
78 | if (rhandle == NULL) | |
79 | return; | |
80 | ||
81 | /* kfree(NULL) is safe */ | |
82 | kfree(rhandle->readdir_response.dirent_array); | |
83 | rhandle->readdir_response.dirent_array = NULL; | |
84 | ||
85 | if (rhandle->buffer_index >= 0) { | |
86 | readdir_index_put(bufmap, rhandle->buffer_index); | |
87 | rhandle->buffer_index = -1; | |
88 | } | |
89 | if (rhandle->dents_buf) { | |
90 | gossip_debug(GOSSIP_DIR_DEBUG, "vfree %p\n", | |
91 | rhandle->dents_buf); | |
92 | vfree(rhandle->dents_buf); | |
93 | rhandle->dents_buf = NULL; | |
94 | } | |
95 | } | |
96 | ||
97 | /* | |
98 | * Read directory entries from an instance of an open directory. | |
5db11c21 MM |
99 | */ |
100 | static int pvfs2_readdir(struct file *file, struct dir_context *ctx) | |
101 | { | |
102 | struct pvfs2_bufmap *bufmap = NULL; | |
103 | int ret = 0; | |
104 | int buffer_index; | |
88309aae MM |
105 | /* |
106 | * ptoken supports Orangefs' distributed directory logic, added | |
107 | * in 2.9.2. | |
108 | */ | |
5db11c21 MM |
109 | __u64 *ptoken = file->private_data; |
110 | __u64 pos = 0; | |
111 | ino_t ino = 0; | |
112 | struct dentry *dentry = file->f_path.dentry; | |
113 | struct pvfs2_kernel_op_s *new_op = NULL; | |
114 | struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(dentry->d_inode); | |
115 | int buffer_full = 0; | |
116 | struct readdir_handle_s rhandle; | |
117 | int i = 0; | |
118 | int len = 0; | |
119 | ino_t current_ino = 0; | |
120 | char *current_entry = NULL; | |
121 | long bytes_decoded; | |
122 | ||
88309aae MM |
123 | gossip_debug(GOSSIP_DIR_DEBUG, |
124 | "%s: ctx->pos:%lld, ptoken = %llu\n", | |
125 | __func__, | |
126 | lld(ctx->pos), | |
127 | llu(*ptoken)); | |
5db11c21 MM |
128 | |
129 | pos = (__u64) ctx->pos; | |
130 | ||
131 | /* are we done? */ | |
132 | if (pos == PVFS_READDIR_END) { | |
133 | gossip_debug(GOSSIP_DIR_DEBUG, | |
134 | "Skipping to termination path\n"); | |
135 | return 0; | |
136 | } | |
137 | ||
138 | gossip_debug(GOSSIP_DIR_DEBUG, | |
139 | "pvfs2_readdir called on %s (pos=%llu)\n", | |
140 | dentry->d_name.name, llu(pos)); | |
141 | ||
142 | rhandle.buffer_index = -1; | |
143 | rhandle.dents_buf = NULL; | |
144 | memset(&rhandle.readdir_response, 0, sizeof(rhandle.readdir_response)); | |
145 | ||
146 | new_op = op_alloc(PVFS2_VFS_OP_READDIR); | |
147 | if (!new_op) | |
148 | return -ENOMEM; | |
149 | ||
150 | new_op->uses_shared_memory = 1; | |
151 | new_op->upcall.req.readdir.refn = pvfs2_inode->refn; | |
152 | new_op->upcall.req.readdir.max_dirent_count = MAX_DIRENT_COUNT_READDIR; | |
153 | ||
154 | gossip_debug(GOSSIP_DIR_DEBUG, | |
155 | "%s: upcall.req.readdir.refn.khandle: %pU\n", | |
156 | __func__, | |
157 | &new_op->upcall.req.readdir.refn.khandle); | |
158 | ||
5db11c21 MM |
159 | new_op->upcall.req.readdir.token = *ptoken; |
160 | ||
161 | get_new_buffer_index: | |
162 | ret = readdir_index_get(&bufmap, &buffer_index); | |
163 | if (ret < 0) { | |
164 | gossip_lerr("pvfs2_readdir: readdir_index_get() failure (%d)\n", | |
165 | ret); | |
166 | goto out_free_op; | |
167 | } | |
168 | new_op->upcall.req.readdir.buf_index = buffer_index; | |
169 | ||
170 | ret = service_operation(new_op, | |
171 | "pvfs2_readdir", | |
172 | get_interruptible_flag(dentry->d_inode)); | |
173 | ||
174 | gossip_debug(GOSSIP_DIR_DEBUG, | |
175 | "Readdir downcall status is %d. ret:%d\n", | |
176 | new_op->downcall.status, | |
177 | ret); | |
178 | ||
179 | if (ret == -EAGAIN && op_state_purged(new_op)) { | |
180 | /* | |
181 | * readdir shared memory aread has been wiped due to | |
182 | * pvfs2-client-core restarting, so we must get a new | |
183 | * index into the shared memory. | |
184 | */ | |
185 | gossip_debug(GOSSIP_DIR_DEBUG, | |
186 | "%s: Getting new buffer_index for retry of readdir..\n", | |
187 | __func__); | |
188 | readdir_index_put(bufmap, buffer_index); | |
189 | goto get_new_buffer_index; | |
190 | } | |
191 | ||
192 | if (ret == -EIO && op_state_purged(new_op)) { | |
193 | gossip_err("%s: Client is down. Aborting readdir call.\n", | |
194 | __func__); | |
195 | readdir_index_put(bufmap, buffer_index); | |
196 | goto out_free_op; | |
197 | } | |
198 | ||
199 | if (ret < 0 || new_op->downcall.status != 0) { | |
200 | gossip_debug(GOSSIP_DIR_DEBUG, | |
201 | "Readdir request failed. Status:%d\n", | |
202 | new_op->downcall.status); | |
203 | readdir_index_put(bufmap, buffer_index); | |
204 | if (ret >= 0) | |
205 | ret = new_op->downcall.status; | |
206 | goto out_free_op; | |
207 | } | |
208 | ||
209 | bytes_decoded = | |
210 | readdir_handle_ctor(&rhandle, | |
211 | new_op->downcall.trailer_buf, | |
212 | buffer_index); | |
213 | if (bytes_decoded < 0) { | |
214 | gossip_err("pvfs2_readdir: Could not decode trailer buffer into a readdir response %d\n", | |
215 | ret); | |
216 | ret = bytes_decoded; | |
217 | readdir_index_put(bufmap, buffer_index); | |
218 | goto out_free_op; | |
219 | } | |
220 | ||
221 | if (bytes_decoded != new_op->downcall.trailer_size) { | |
88309aae MM |
222 | gossip_err("pvfs2_readdir: # bytes decoded (%ld) " |
223 | "!= trailer size (%ld)\n", | |
224 | bytes_decoded, | |
225 | (long)new_op->downcall.trailer_size); | |
5db11c21 MM |
226 | ret = -EINVAL; |
227 | goto out_destroy_handle; | |
228 | } | |
229 | ||
88309aae MM |
230 | /* |
231 | * pvfs2 doesn't actually store dot and dot-dot, but | |
232 | * we need to have them represented. | |
233 | */ | |
5db11c21 MM |
234 | if (pos == 0) { |
235 | ino = get_ino_from_khandle(dentry->d_inode); | |
236 | gossip_debug(GOSSIP_DIR_DEBUG, | |
237 | "%s: calling dir_emit of \".\" with pos = %llu\n", | |
238 | __func__, | |
239 | llu(pos)); | |
240 | ret = dir_emit(ctx, ".", 1, ino, DT_DIR); | |
88309aae | 241 | pos += 1; |
5db11c21 MM |
242 | } |
243 | ||
244 | if (pos == 1) { | |
245 | ino = get_parent_ino_from_dentry(dentry); | |
246 | gossip_debug(GOSSIP_DIR_DEBUG, | |
247 | "%s: calling dir_emit of \"..\" with pos = %llu\n", | |
248 | __func__, | |
249 | llu(pos)); | |
250 | ret = dir_emit(ctx, "..", 2, ino, DT_DIR); | |
88309aae | 251 | pos += 1; |
5db11c21 MM |
252 | } |
253 | ||
88309aae MM |
254 | /* |
255 | * we stored PVFS_ITERATE_NEXT in ctx->pos last time around | |
256 | * to prevent "finding" dot and dot-dot on any iteration | |
257 | * other than the first. | |
258 | */ | |
259 | if (ctx->pos == PVFS_ITERATE_NEXT) | |
260 | ctx->pos = 0; | |
261 | ||
262 | for (i = ctx->pos; | |
263 | i < rhandle.readdir_response.pvfs_dirent_outcount; | |
264 | i++) { | |
5db11c21 MM |
265 | len = rhandle.readdir_response.dirent_array[i].d_length; |
266 | current_entry = rhandle.readdir_response.dirent_array[i].d_name; | |
267 | current_ino = pvfs2_khandle_to_ino( | |
268 | &(rhandle.readdir_response.dirent_array[i].khandle)); | |
269 | ||
270 | gossip_debug(GOSSIP_DIR_DEBUG, | |
88309aae MM |
271 | "calling dir_emit for %s with len %d" |
272 | ", ctx->pos %ld\n", | |
5db11c21 MM |
273 | current_entry, |
274 | len, | |
88309aae MM |
275 | (unsigned long)ctx->pos); |
276 | /* | |
277 | * type is unknown. We don't return object type | |
278 | * in the dirent_array. This leaves getdents | |
279 | * clueless about type. | |
280 | */ | |
5db11c21 MM |
281 | ret = |
282 | dir_emit(ctx, current_entry, len, current_ino, DT_UNKNOWN); | |
88309aae MM |
283 | if (!ret) |
284 | break; | |
5db11c21 | 285 | ctx->pos++; |
88309aae | 286 | gossip_debug(GOSSIP_DIR_DEBUG, |
5db11c21 MM |
287 | "%s: ctx->pos:%lld\n", |
288 | __func__, | |
289 | lld(ctx->pos)); | |
290 | ||
5db11c21 MM |
291 | } |
292 | ||
54804949 | 293 | /* |
88309aae MM |
294 | * we ran all the way through the last batch, set up for |
295 | * getting another batch... | |
296 | */ | |
297 | if (ret) { | |
5db11c21 | 298 | *ptoken = rhandle.readdir_response.token; |
88309aae | 299 | ctx->pos = PVFS_ITERATE_NEXT; |
5db11c21 MM |
300 | } |
301 | ||
302 | /* | |
303 | * Did we hit the end of the directory? | |
304 | */ | |
305 | if (rhandle.readdir_response.token == PVFS_READDIR_END && | |
306 | !buffer_full) { | |
88309aae MM |
307 | gossip_debug(GOSSIP_DIR_DEBUG, |
308 | "End of dir detected; setting ctx->pos to PVFS_READDIR_END.\n"); | |
5db11c21 MM |
309 | ctx->pos = PVFS_READDIR_END; |
310 | } | |
311 | ||
5db11c21 MM |
312 | out_destroy_handle: |
313 | readdir_handle_dtor(bufmap, &rhandle); | |
314 | out_free_op: | |
315 | op_release(new_op); | |
316 | gossip_debug(GOSSIP_DIR_DEBUG, "pvfs2_readdir returning %d\n", ret); | |
317 | return ret; | |
318 | } | |
319 | ||
320 | static int pvfs2_dir_open(struct inode *inode, struct file *file) | |
321 | { | |
322 | __u64 *ptoken; | |
323 | ||
324 | file->private_data = kmalloc(sizeof(__u64), GFP_KERNEL); | |
325 | if (!file->private_data) | |
326 | return -ENOMEM; | |
327 | ||
328 | ptoken = file->private_data; | |
329 | *ptoken = PVFS_READDIR_START; | |
330 | return 0; | |
331 | } | |
332 | ||
333 | static int pvfs2_dir_release(struct inode *inode, struct file *file) | |
334 | { | |
335 | pvfs2_flush_inode(inode); | |
336 | kfree(file->private_data); | |
337 | return 0; | |
338 | } | |
339 | ||
340 | /** PVFS2 implementation of VFS directory operations */ | |
341 | const struct file_operations pvfs2_dir_operations = { | |
342 | .read = generic_read_dir, | |
343 | .iterate = pvfs2_readdir, | |
344 | .open = pvfs2_dir_open, | |
345 | .release = pvfs2_dir_release, | |
346 | }; |