| 1 |
/* |
| 2 |
Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> |
| 3 |
This file is part of GlusterFS. |
| 4 |
|
| 5 |
This file is licensed to you under your choice of the GNU Lesser |
| 6 |
General Public License, version 3 or any later version (LGPLv3 or |
| 7 |
later), or the GNU General Public License, version 2 (GPLv2), in all |
| 8 |
cases as published by the Free Software Foundation. |
| 9 |
*/ |
| 10 |
|
| 11 |
#include <libgen.h> |
| 12 |
#include <unistd.h> |
| 13 |
#include <sys/time.h> |
| 14 |
#include <stdlib.h> |
| 15 |
#include <signal.h> |
| 16 |
#include <string.h> |
| 17 |
|
| 18 |
#include <glusterfs/dict.h> |
| 19 |
#include <glusterfs/list.h> |
| 20 |
#include <glusterfs/compat-errno.h> |
| 21 |
#include <glusterfs/compat.h> |
| 22 |
#include "afr-transaction.h" |
| 23 |
|
| 24 |
static int32_t |
| 25 |
afr_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, |
| 26 |
int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata) |
| 27 |
{ |
| 28 |
afr_local_t *local = NULL; |
| 29 |
int call_count = -1; |
| 30 |
int32_t child_index = 0; |
| 31 |
afr_fd_ctx_t *fd_ctx = NULL; |
| 32 |
|
| 33 |
local = frame->local; |
| 34 |
fd_ctx = local->fd_ctx; |
| 35 |
child_index = (long)cookie; |
| 36 |
|
| 37 |
local->replies[child_index].valid = 1; |
| 38 |
local->replies[child_index].op_ret = op_ret; |
| 39 |
local->replies[child_index].op_errno = op_errno; |
| 40 |
|
| 41 |
LOCK(&frame->lock); |
| 42 |
{ |
| 43 |
if (op_ret == -1) { |
| 44 |
local->op_errno = op_errno; |
| 45 |
fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED; |
| 46 |
} else { |
| 47 |
local->op_ret = op_ret; |
| 48 |
fd_ctx->opened_on[child_index] = AFR_FD_OPENED; |
| 49 |
if (!local->xdata_rsp && xdata) |
| 50 |
local->xdata_rsp = dict_ref(xdata); |
| 51 |
} |
| 52 |
call_count = --local->call_count; |
| 53 |
} |
| 54 |
UNLOCK(&frame->lock); |
| 55 |
|
| 56 |
if (call_count == 0) { |
| 57 |
afr_handle_replies_quorum(frame, this); |
| 58 |
AFR_STACK_UNWIND(opendir, frame, local->op_ret, local->op_errno, |
| 59 |
local->fd, NULL); |
| 60 |
} |
| 61 |
|
| 62 |
return 0; |
| 63 |
} |
| 64 |
|
| 65 |
int |
| 66 |
afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, |
| 67 |
dict_t *xdata) |
| 68 |
{ |
| 69 |
afr_private_t *priv = NULL; |
| 70 |
afr_local_t *local = NULL; |
| 71 |
int i = 0; |
| 72 |
int call_count = -1; |
| 73 |
int32_t op_errno = ENOMEM; |
| 74 |
afr_fd_ctx_t *fd_ctx = NULL; |
| 75 |
|
| 76 |
priv = this->private; |
| 77 |
|
| 78 |
local = AFR_FRAME_INIT(frame, op_errno); |
| 79 |
if (!local) |
| 80 |
goto out; |
| 81 |
|
| 82 |
local->op = GF_FOP_OPENDIR; |
| 83 |
|
| 84 |
if (priv->quorum_count && !afr_has_quorum(local->child_up, priv, NULL)) { |
| 85 |
op_errno = afr_quorum_errno(priv); |
| 86 |
goto out; |
| 87 |
} |
| 88 |
|
| 89 |
if (!afr_is_consistent_io_possible(local, priv, &op_errno)) |
| 90 |
goto out; |
| 91 |
|
| 92 |
fd_ctx = afr_fd_ctx_get(fd, this); |
| 93 |
if (!fd_ctx) |
| 94 |
goto out; |
| 95 |
|
| 96 |
loc_copy(&local->loc, loc); |
| 97 |
|
| 98 |
local->fd = fd_ref(fd); |
| 99 |
local->fd_ctx = fd_ctx; |
| 100 |
|
| 101 |
call_count = local->call_count; |
| 102 |
|
| 103 |
for (i = 0; i < priv->child_count; i++) { |
| 104 |
if (local->child_up[i]) { |
| 105 |
STACK_WIND_COOKIE(frame, afr_opendir_cbk, (void *)(long)i, |
| 106 |
priv->children[i], |
| 107 |
priv->children[i]->fops->opendir, loc, fd, NULL); |
| 108 |
|
| 109 |
if (!--call_count) |
| 110 |
break; |
| 111 |
} |
| 112 |
} |
| 113 |
|
| 114 |
return 0; |
| 115 |
out: |
| 116 |
AFR_STACK_UNWIND(opendir, frame, -1, op_errno, fd, NULL); |
| 117 |
return 0; |
| 118 |
} |
| 119 |
|
| 120 |
static int |
| 121 |
afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol) |
| 122 |
{ |
| 123 |
int gen = 0; |
| 124 |
int entry_read_subvol = 0; |
| 125 |
unsigned char *data_readable = NULL; |
| 126 |
unsigned char *metadata_readable = NULL; |
| 127 |
afr_private_t *priv = NULL; |
| 128 |
|
| 129 |
priv = this->private; |
| 130 |
data_readable = alloca0(priv->child_count); |
| 131 |
metadata_readable = alloca0(priv->child_count); |
| 132 |
|
| 133 |
afr_inode_read_subvol_get(inode, this, data_readable, metadata_readable, |
| 134 |
&gen); |
| 135 |
|
| 136 |
if (gen != priv->event_generation || !data_readable[par_read_subvol] || |
| 137 |
!metadata_readable[par_read_subvol]) |
| 138 |
return -1; |
| 139 |
|
| 140 |
/* Once the control reaches the following statement, it means that the |
| 141 |
* parent's read subvol is perfectly readable. So calling |
| 142 |
* either afr_data_subvol_get() or afr_metadata_subvol_get() would |
| 143 |
* yield the same result. Hence, choosing afr_data_subvol_get() below. |
| 144 |
*/ |
| 145 |
|
| 146 |
if (!priv->consistent_metadata) |
| 147 |
return 0; |
| 148 |
|
| 149 |
/* For an inode fetched through readdirp which is yet to be linked, |
| 150 |
* inode ctx would not be initialised (yet). So this function returns |
| 151 |
* -1 above due to gen being 0, which is why it is OK to pass NULL for |
| 152 |
* read_subvol_args here. |
| 153 |
*/ |
| 154 |
entry_read_subvol = afr_data_subvol_get(inode, this, NULL, NULL, NULL, |
| 155 |
NULL); |
| 156 |
if (entry_read_subvol != par_read_subvol) |
| 157 |
return -1; |
| 158 |
|
| 159 |
return 0; |
| 160 |
} |
| 161 |
|
| 162 |
static int32_t |
| 163 |
afr_readdir_transform_entries(pid_t pid, xlator_t *this, |
| 164 |
gf_dirent_t *subvol_entries, int subvol, |
| 165 |
gf_dirent_t *entries, fd_t *fd) |
| 166 |
{ |
| 167 |
int ret; |
| 168 |
gf_dirent_t *entry = NULL; |
| 169 |
gf_dirent_t *tmp = NULL; |
| 170 |
afr_private_t *priv = NULL; |
| 171 |
gf_boolean_t need_heal = _gf_false; |
| 172 |
gf_boolean_t validate_subvol = _gf_false; |
| 173 |
int32_t count = 0; |
| 174 |
gf_boolean_t is_root_gfid = __is_root_gfid(fd->inode->gfid); |
| 175 |
|
| 176 |
priv = this->private; |
| 177 |
|
| 178 |
need_heal = afr_get_need_heal(priv); |
| 179 |
validate_subvol = need_heal | priv->consistent_metadata; |
| 180 |
|
| 181 |
list_for_each_entry_safe(entry, tmp, &subvol_entries->list, list) |
| 182 |
{ |
| 183 |
/* we can skip afr_is_private_directory() check unless it's root gfid |
| 184 |
* which we check once (above) for all entries |
| 185 |
*/ |
| 186 |
if (is_root_gfid) { |
| 187 |
if (afr_is_private_directory(priv, entry->d_name, pid)) { |
| 188 |
continue; |
| 189 |
} |
| 190 |
} |
| 191 |
|
| 192 |
list_move_tail(&entry->list, &entries->list); |
| 193 |
count++; |
| 194 |
|
| 195 |
if (!validate_subvol) |
| 196 |
continue; |
| 197 |
|
| 198 |
if (entry->inode) { |
| 199 |
ret = afr_validate_read_subvol(entry->inode, this, subvol); |
| 200 |
if (ret != 0) { |
| 201 |
inode_unref(entry->inode); |
| 202 |
entry->inode = NULL; |
| 203 |
continue; |
| 204 |
} |
| 205 |
} |
| 206 |
} |
| 207 |
|
| 208 |
return count; |
| 209 |
} |
| 210 |
|
| 211 |
static int32_t |
| 212 |
afr_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, |
| 213 |
int32_t op_ret, int32_t op_errno, gf_dirent_t *subvol_entries, |
| 214 |
dict_t *xdata) |
| 215 |
{ |
| 216 |
afr_local_t *local = NULL; |
| 217 |
gf_dirent_t entries; |
| 218 |
|
| 219 |
INIT_LIST_HEAD(&entries.list); |
| 220 |
|
| 221 |
local = frame->local; |
| 222 |
|
| 223 |
if (op_ret < 0 && !local->cont.readdir.offset) { |
| 224 |
/* failover only if this was first readdir, detected |
| 225 |
by offset == 0 */ |
| 226 |
local->op_ret = op_ret; |
| 227 |
local->op_errno = op_errno; |
| 228 |
|
| 229 |
afr_read_txn_continue(frame, this, (long)cookie); |
| 230 |
return 0; |
| 231 |
} |
| 232 |
|
| 233 |
if (op_ret >= 0) |
| 234 |
op_ret = afr_readdir_transform_entries(frame->root->pid, this, |
| 235 |
subvol_entries, (long)cookie, |
| 236 |
&entries, local->fd); |
| 237 |
|
| 238 |
AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata); |
| 239 |
|
| 240 |
gf_dirent_free(&entries); |
| 241 |
|
| 242 |
return 0; |
| 243 |
} |
| 244 |
|
| 245 |
static int |
| 246 |
afr_readdir_wind(call_frame_t *frame, xlator_t *this, int subvol) |
| 247 |
{ |
| 248 |
afr_local_t *local = NULL; |
| 249 |
afr_private_t *priv = NULL; |
| 250 |
afr_fd_ctx_t *fd_ctx = NULL; |
| 251 |
|
| 252 |
priv = this->private; |
| 253 |
local = frame->local; |
| 254 |
fd_ctx = afr_fd_ctx_get(local->fd, this); |
| 255 |
if (!fd_ctx) { |
| 256 |
local->op_errno = EINVAL; |
| 257 |
local->op_ret = -1; |
| 258 |
} |
| 259 |
|
| 260 |
if (subvol == -1 || !fd_ctx) { |
| 261 |
AFR_STACK_UNWIND(readdir, frame, local->op_ret, local->op_errno, 0, 0); |
| 262 |
return 0; |
| 263 |
} |
| 264 |
|
| 265 |
fd_ctx->readdir_subvol = subvol; |
| 266 |
|
| 267 |
if (local->op == GF_FOP_READDIR) |
| 268 |
STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol, |
| 269 |
priv->children[subvol], |
| 270 |
priv->children[subvol]->fops->readdir, local->fd, |
| 271 |
local->cont.readdir.size, local->cont.readdir.offset, |
| 272 |
local->xdata_req); |
| 273 |
else |
| 274 |
STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol, |
| 275 |
priv->children[subvol], |
| 276 |
priv->children[subvol]->fops->readdirp, local->fd, |
| 277 |
local->cont.readdir.size, local->cont.readdir.offset, |
| 278 |
local->xdata_req); |
| 279 |
return 0; |
| 280 |
} |
| 281 |
|
| 282 |
static int |
| 283 |
afr_do_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, |
| 284 |
off_t offset, int whichop, dict_t *dict) |
| 285 |
{ |
| 286 |
afr_local_t *local = NULL; |
| 287 |
int32_t op_errno = 0; |
| 288 |
int subvol = -1; |
| 289 |
afr_fd_ctx_t *fd_ctx = NULL; |
| 290 |
|
| 291 |
local = AFR_FRAME_INIT(frame, op_errno); |
| 292 |
if (!local) |
| 293 |
goto out; |
| 294 |
|
| 295 |
fd_ctx = afr_fd_ctx_get(fd, this); |
| 296 |
if (!fd_ctx) { |
| 297 |
op_errno = EINVAL; |
| 298 |
goto out; |
| 299 |
} |
| 300 |
|
| 301 |
local->op = whichop; |
| 302 |
local->fd = fd_ref(fd); |
| 303 |
local->cont.readdir.size = size; |
| 304 |
local->cont.readdir.offset = offset; |
| 305 |
local->xdata_req = (dict) ? dict_ref(dict) : NULL; |
| 306 |
|
| 307 |
subvol = fd_ctx->readdir_subvol; |
| 308 |
|
| 309 |
if (offset == 0 || subvol == -1) { |
| 310 |
/* First readdir has option of failing over and selecting |
| 311 |
an appropriate read subvolume */ |
| 312 |
afr_read_txn(frame, this, fd->inode, afr_readdir_wind, |
| 313 |
AFR_DATA_TRANSACTION); |
| 314 |
} else { |
| 315 |
/* But continued readdirs MUST stick to the same subvolume |
| 316 |
without an option to failover */ |
| 317 |
afr_readdir_wind(frame, this, subvol); |
| 318 |
} |
| 319 |
|
| 320 |
return 0; |
| 321 |
out: |
| 322 |
AFR_STACK_UNWIND(readdir, frame, -1, op_errno, NULL, NULL); |
| 323 |
return 0; |
| 324 |
} |
| 325 |
|
| 326 |
int32_t |
| 327 |
afr_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, |
| 328 |
off_t offset, dict_t *xdata) |
| 329 |
{ |
| 330 |
afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIR, xdata); |
| 331 |
|
| 332 |
return 0; |
| 333 |
} |
| 334 |
|
| 335 |
int32_t |
| 336 |
afr_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, |
| 337 |
off_t offset, dict_t *dict) |
| 338 |
{ |
| 339 |
afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIRP, dict); |
| 340 |
|
| 341 |
return 0; |
| 342 |
} |
| 343 |
|
| 344 |
int32_t |
| 345 |
afr_releasedir(xlator_t *this, fd_t *fd) |
| 346 |
{ |
| 347 |
afr_cleanup_fd_ctx(this, fd); |
| 348 |
|
| 349 |
return 0; |
| 350 |
} |