| 1 | /* |
| 2 | * 9p backend |
| 3 | * |
| 4 | * Copyright IBM, Corp. 2011 |
| 5 | * |
| 6 | * Authors: |
| 7 | * Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> |
| 8 | * |
| 9 | * This work is licensed under the terms of the GNU GPL, version 2. See |
| 10 | * the COPYING file in the top-level directory. |
| 11 | * |
| 12 | */ |
| 13 | |
| 14 | /* |
| 15 | * Not so fast! You might want to read the 9p developer docs first: |
| 16 | * https://wiki.qemu.org/Documentation/9p |
| 17 | */ |
| 18 | |
| 19 | #include "qemu/osdep.h" |
| 20 | #include "fsdev/qemu-fsdev.h" |
| 21 | #include "qemu/thread.h" |
| 22 | #include "qemu/main-loop.h" |
| 23 | #include "qemu/error-report.h" |
| 24 | #include "coth.h" |
| 25 | #include "9p-xattr.h" |
| 26 | #include "9p-util.h" |
| 27 | |
| 28 | /* |
| 29 | * Intended to be called from bottom-half (e.g. background I/O thread) |
| 30 | * context. |
| 31 | */ |
| 32 | static int do_readdir(V9fsPDU *pdu, V9fsFidState *fidp, struct dirent **dent) |
| 33 | { |
| 34 | int err = 0; |
| 35 | V9fsState *s = pdu->s; |
| 36 | struct dirent *entry; |
| 37 | |
| 38 | errno = 0; |
| 39 | entry = s->ops->readdir(&s->ctx, &fidp->fs); |
| 40 | if (!entry && errno) { |
| 41 | *dent = NULL; |
| 42 | err = -errno; |
| 43 | } else { |
| 44 | *dent = entry; |
| 45 | } |
| 46 | return err; |
| 47 | } |
| 48 | |
| 49 | /* |
| 50 | * TODO: This will be removed for performance reasons. |
| 51 | * Use v9fs_co_readdir_many() instead. |
| 52 | */ |
| 53 | int coroutine_fn v9fs_co_readdir(V9fsPDU *pdu, V9fsFidState *fidp, |
| 54 | struct dirent **dent) |
| 55 | { |
| 56 | int err; |
| 57 | |
| 58 | if (v9fs_request_cancelled(pdu)) { |
| 59 | return -EINTR; |
| 60 | } |
| 61 | v9fs_co_run_in_worker({ |
| 62 | err = do_readdir(pdu, fidp, dent); |
| 63 | }); |
| 64 | return err; |
| 65 | } |
| 66 | |
| 67 | /* |
| 68 | * This is solely executed on a background IO thread. |
| 69 | * |
| 70 | * See v9fs_co_readdir_many() (as its only user) below for details. |
| 71 | */ |
| 72 | static int coroutine_fn |
| 73 | do_readdir_many(V9fsPDU *pdu, V9fsFidState *fidp, struct V9fsDirEnt **entries, |
| 74 | off_t offset, int32_t maxsize, bool dostat) |
| 75 | { |
| 76 | V9fsState *s = pdu->s; |
| 77 | V9fsString name; |
| 78 | int len, err = 0; |
| 79 | int32_t size = 0; |
| 80 | off_t saved_dir_pos; |
| 81 | struct dirent *dent; |
| 82 | struct V9fsDirEnt *e = NULL; |
| 83 | V9fsPath path; |
| 84 | struct stat stbuf; |
| 85 | |
| 86 | *entries = NULL; |
| 87 | v9fs_path_init(&path); |
| 88 | |
| 89 | /* |
| 90 | * TODO: Here should be a warn_report_once() if lock failed. |
| 91 | * |
| 92 | * With a good 9p client we should not get into concurrency here, |
| 93 | * because a good client would not use the same fid for concurrent |
| 94 | * requests. We do the lock here for safety reasons though. However |
| 95 | * the client would then suffer performance issues, so better log that |
| 96 | * issue here. |
| 97 | */ |
| 98 | v9fs_readdir_lock(&fidp->fs.dir); |
| 99 | |
| 100 | /* seek directory to requested initial position */ |
| 101 | if (offset == 0) { |
| 102 | s->ops->rewinddir(&s->ctx, &fidp->fs); |
| 103 | } else { |
| 104 | s->ops->seekdir(&s->ctx, &fidp->fs, offset); |
| 105 | } |
| 106 | |
| 107 | /* save the directory position */ |
| 108 | saved_dir_pos = s->ops->telldir(&s->ctx, &fidp->fs); |
| 109 | if (saved_dir_pos < 0) { |
| 110 | err = saved_dir_pos; |
| 111 | goto out; |
| 112 | } |
| 113 | |
| 114 | while (true) { |
| 115 | /* interrupt loop if request was cancelled by a Tflush request */ |
| 116 | if (v9fs_request_cancelled(pdu)) { |
| 117 | err = -EINTR; |
| 118 | break; |
| 119 | } |
| 120 | |
| 121 | /* get directory entry from fs driver */ |
| 122 | err = do_readdir(pdu, fidp, &dent); |
| 123 | if (err || !dent) { |
| 124 | break; |
| 125 | } |
| 126 | |
| 127 | /* |
| 128 | * stop this loop as soon as it would exceed the allowed maximum |
| 129 | * response message size for the directory entries collected so far, |
| 130 | * because anything beyond that size would need to be discarded by |
| 131 | * 9p controller (main thread / top half) anyway |
| 132 | */ |
| 133 | v9fs_string_init(&name); |
| 134 | v9fs_string_sprintf(&name, "%s", dent->d_name); |
| 135 | len = v9fs_readdir_response_size(&name); |
| 136 | v9fs_string_free(&name); |
| 137 | if (size + len > maxsize) { |
| 138 | /* this is not an error case actually */ |
| 139 | break; |
| 140 | } |
| 141 | |
| 142 | /* append next node to result chain */ |
| 143 | if (!e) { |
| 144 | *entries = e = g_new0(V9fsDirEnt, 1); |
| 145 | } else { |
| 146 | e = e->next = g_new0(V9fsDirEnt, 1); |
| 147 | } |
| 148 | e->dent = qemu_dirent_dup(dent); |
| 149 | |
| 150 | /* perform a full stat() for directory entry if requested by caller */ |
| 151 | if (dostat) { |
| 152 | err = s->ops->name_to_path( |
| 153 | &s->ctx, &fidp->path, dent->d_name, &path |
| 154 | ); |
| 155 | if (err < 0) { |
| 156 | err = -errno; |
| 157 | break; |
| 158 | } |
| 159 | |
| 160 | err = s->ops->lstat(&s->ctx, &path, &stbuf); |
| 161 | if (err < 0) { |
| 162 | err = -errno; |
| 163 | break; |
| 164 | } |
| 165 | |
| 166 | e->st = g_new0(struct stat, 1); |
| 167 | memcpy(e->st, &stbuf, sizeof(struct stat)); |
| 168 | } |
| 169 | |
| 170 | size += len; |
| 171 | saved_dir_pos = qemu_dirent_off(dent); |
| 172 | } |
| 173 | |
| 174 | /* restore (last) saved position */ |
| 175 | s->ops->seekdir(&s->ctx, &fidp->fs, saved_dir_pos); |
| 176 | |
| 177 | out: |
| 178 | v9fs_readdir_unlock(&fidp->fs.dir); |
| 179 | v9fs_path_free(&path); |
| 180 | if (err < 0) { |
| 181 | return err; |
| 182 | } |
| 183 | return size; |
| 184 | } |
| 185 | |
| 186 | /** |
| 187 | * v9fs_co_readdir_many() - Reads multiple directory entries in one rush. |
| 188 | * |
| 189 | * @pdu: the causing 9p (T_readdir) client request |
| 190 | * @fidp: already opened directory where readdir shall be performed on |
| 191 | * @entries: output for directory entries (must not be NULL) |
| 192 | * @offset: initial position inside the directory the function shall |
| 193 | * seek to before retrieving the directory entries |
| 194 | * @maxsize: maximum result message body size (in bytes) |
| 195 | * @dostat: whether a stat() should be performed and returned for |
| 196 | * each directory entry |
| 197 | * Return: resulting response message body size (in bytes) on success, |
| 198 | * negative error code otherwise |
| 199 | * |
| 200 | * Retrieves the requested (max. amount of) directory entries from the fs |
| 201 | * driver. This function must only be called by the main IO thread (top half). |
| 202 | * Internally this function call will be dispatched to a background IO thread |
| 203 | * (bottom half) where it is eventually executed by the fs driver. |
| 204 | * |
| 205 | * Acquiring multiple directory entries in one rush from the fs |
| 206 | * driver, instead of retrieving each directory entry individually, is very |
| 207 | * beneficial from performance point of view. Because for every fs driver |
| 208 | * request latency is added, which in practice could lead to overall |
| 209 | * latencies of several hundred ms for reading all entries (of just a single |
| 210 | * directory) if every directory entry was individually requested from fs |
| 211 | * driver. |
| 212 | * |
| 213 | * NOTE: You must ALWAYS call v9fs_free_dirents(entries) after calling |
| 214 | * v9fs_co_readdir_many(), both on success and on error cases of this |
| 215 | * function, to avoid memory leaks once @entries are no longer needed. |
| 216 | */ |
| 217 | int coroutine_fn v9fs_co_readdir_many(V9fsPDU *pdu, V9fsFidState *fidp, |
| 218 | struct V9fsDirEnt **entries, |
| 219 | off_t offset, int32_t maxsize, |
| 220 | bool dostat) |
| 221 | { |
| 222 | int err = 0; |
| 223 | V9fsState *s = pdu->s; |
| 224 | |
| 225 | if (v9fs_request_cancelled(pdu)) { |
| 226 | return -EINTR; |
| 227 | } |
| 228 | v9fs_path_read_lock(s); |
| 229 | v9fs_co_run_in_worker({ |
| 230 | err = do_readdir_many(pdu, fidp, entries, offset, maxsize, dostat); |
| 231 | }); |
| 232 | v9fs_path_unlock(s); |
| 233 | return err; |
| 234 | } |
| 235 | |
| 236 | off_t v9fs_co_telldir(V9fsPDU *pdu, V9fsFidState *fidp) |
| 237 | { |
| 238 | off_t err; |
| 239 | V9fsState *s = pdu->s; |
| 240 | |
| 241 | if (v9fs_request_cancelled(pdu)) { |
| 242 | return -EINTR; |
| 243 | } |
| 244 | v9fs_co_run_in_worker( |
| 245 | { |
| 246 | err = s->ops->telldir(&s->ctx, &fidp->fs); |
| 247 | if (err < 0) { |
| 248 | err = -errno; |
| 249 | } |
| 250 | }); |
| 251 | return err; |
| 252 | } |
| 253 | |
| 254 | void coroutine_fn v9fs_co_seekdir(V9fsPDU *pdu, V9fsFidState *fidp, |
| 255 | off_t offset) |
| 256 | { |
| 257 | V9fsState *s = pdu->s; |
| 258 | if (v9fs_request_cancelled(pdu)) { |
| 259 | return; |
| 260 | } |
| 261 | v9fs_co_run_in_worker( |
| 262 | { |
| 263 | s->ops->seekdir(&s->ctx, &fidp->fs, offset); |
| 264 | }); |
| 265 | } |
| 266 | |
| 267 | void coroutine_fn v9fs_co_rewinddir(V9fsPDU *pdu, V9fsFidState *fidp) |
| 268 | { |
| 269 | V9fsState *s = pdu->s; |
| 270 | if (v9fs_request_cancelled(pdu)) { |
| 271 | return; |
| 272 | } |
| 273 | v9fs_co_run_in_worker( |
| 274 | { |
| 275 | s->ops->rewinddir(&s->ctx, &fidp->fs); |
| 276 | }); |
| 277 | } |
| 278 | |
| 279 | int coroutine_fn v9fs_co_mkdir(V9fsPDU *pdu, V9fsFidState *fidp, |
| 280 | V9fsString *name, mode_t mode, uid_t uid, |
| 281 | gid_t gid, struct stat *stbuf) |
| 282 | { |
| 283 | int err; |
| 284 | FsCred cred; |
| 285 | V9fsPath path; |
| 286 | V9fsState *s = pdu->s; |
| 287 | |
| 288 | if (v9fs_request_cancelled(pdu)) { |
| 289 | return -EINTR; |
| 290 | } |
| 291 | cred_init(&cred); |
| 292 | cred.fc_mode = mode; |
| 293 | cred.fc_uid = uid; |
| 294 | cred.fc_gid = gid; |
| 295 | v9fs_path_read_lock(s); |
| 296 | v9fs_co_run_in_worker( |
| 297 | { |
| 298 | err = s->ops->mkdir(&s->ctx, &fidp->path, name->data, &cred); |
| 299 | if (err < 0) { |
| 300 | err = -errno; |
| 301 | } else { |
| 302 | v9fs_path_init(&path); |
| 303 | err = v9fs_name_to_path(s, &fidp->path, name->data, &path); |
| 304 | if (!err) { |
| 305 | err = s->ops->lstat(&s->ctx, &path, stbuf); |
| 306 | if (err < 0) { |
| 307 | err = -errno; |
| 308 | } |
| 309 | } |
| 310 | v9fs_path_free(&path); |
| 311 | } |
| 312 | }); |
| 313 | v9fs_path_unlock(s); |
| 314 | return err; |
| 315 | } |
| 316 | |
| 317 | int coroutine_fn v9fs_co_opendir(V9fsPDU *pdu, V9fsFidState *fidp) |
| 318 | { |
| 319 | int err; |
| 320 | V9fsState *s = pdu->s; |
| 321 | |
| 322 | if (v9fs_request_cancelled(pdu)) { |
| 323 | return -EINTR; |
| 324 | } |
| 325 | v9fs_path_read_lock(s); |
| 326 | v9fs_co_run_in_worker( |
| 327 | { |
| 328 | err = s->ops->opendir(&s->ctx, &fidp->path, &fidp->fs); |
| 329 | if (err < 0) { |
| 330 | err = -errno; |
| 331 | } else { |
| 332 | err = 0; |
| 333 | } |
| 334 | }); |
| 335 | v9fs_path_unlock(s); |
| 336 | if (!err) { |
| 337 | total_open_fd++; |
| 338 | if (total_open_fd > open_fd_hw) { |
| 339 | v9fs_reclaim_fd(pdu); |
| 340 | } |
| 341 | } |
| 342 | return err; |
| 343 | } |
| 344 | |
| 345 | int coroutine_fn v9fs_co_closedir(V9fsPDU *pdu, V9fsFidOpenState *fs) |
| 346 | { |
| 347 | int err; |
| 348 | V9fsState *s = pdu->s; |
| 349 | |
| 350 | if (v9fs_request_cancelled(pdu)) { |
| 351 | return -EINTR; |
| 352 | } |
| 353 | v9fs_co_run_in_worker( |
| 354 | { |
| 355 | err = s->ops->closedir(&s->ctx, fs); |
| 356 | if (err < 0) { |
| 357 | err = -errno; |
| 358 | } |
| 359 | }); |
| 360 | /* 'man 2 close' suggests to ignore close() errors except of EBADF */ |
| 361 | if (unlikely(err && errno == EBADF)) { |
| 362 | /* unexpected case as we should have checked for a valid file handle */ |
| 363 | error_report("9pfs: WARNING: v9fs_co_closedir() failed with EBADF"); |
| 364 | } else { |
| 365 | total_open_fd--; |
| 366 | } |
| 367 | return err; |
| 368 | } |