| 1 |
// SPDX-License-Identifier: GPL-2.0-or-later |
| 2 |
/* Fileserver-directed operation handling. |
| 3 |
* |
| 4 |
* Copyright (C) 2020 Red Hat, Inc. All Rights Reserved. |
| 5 |
* Written by David Howells ([email protected]) |
| 6 |
*/ |
| 7 |
|
| 8 |
#include <linux/kernel.h> |
| 9 |
#include <linux/slab.h> |
| 10 |
#include <linux/fs.h> |
| 11 |
#include "internal.h" |
| 12 |
|
| 13 |
static atomic_t afs_operation_debug_counter; |
| 14 |
|
| 15 |
/* |
| 16 |
* Create an operation against a volume. |
| 17 |
*/ |
| 18 |
struct afs_operation *afs_alloc_operation(struct key *key, struct afs_volume *volume) |
| 19 |
{ |
| 20 |
struct afs_operation *op; |
| 21 |
|
| 22 |
_enter(""); |
| 23 |
|
| 24 |
op = kzalloc_obj(*op); |
| 25 |
if (!op) |
| 26 |
return ERR_PTR(-ENOMEM); |
| 27 |
|
| 28 |
if (!key) { |
| 29 |
key = afs_request_key(volume->cell); |
| 30 |
if (IS_ERR(key)) { |
| 31 |
kfree(op); |
| 32 |
return ERR_CAST(key); |
| 33 |
} |
| 34 |
} else { |
| 35 |
key_get(key); |
| 36 |
} |
| 37 |
|
| 38 |
op->key = key; |
| 39 |
op->volume = afs_get_volume(volume, afs_volume_trace_get_new_op); |
| 40 |
op->net = volume->cell->net; |
| 41 |
op->cb_v_break = atomic_read(&volume->cb_v_break); |
| 42 |
op->pre_volsync.creation = volume->creation_time; |
| 43 |
op->pre_volsync.update = volume->update_time; |
| 44 |
op->debug_id = atomic_inc_return(&afs_operation_debug_counter); |
| 45 |
op->nr_iterations = -1; |
| 46 |
afs_op_set_error(op, -EDESTADDRREQ); |
| 47 |
|
| 48 |
_leave(" = [op=%08x]", op->debug_id); |
| 49 |
return op; |
| 50 |
} |
| 51 |
|
| 52 |
struct afs_io_locker { |
| 53 |
struct list_head link; |
| 54 |
struct task_struct *task; |
| 55 |
unsigned long have_lock; |
| 56 |
}; |
| 57 |
|
| 58 |
/* |
| 59 |
* Unlock the I/O lock on a vnode. |
| 60 |
*/ |
| 61 |
static void afs_unlock_for_io(struct afs_vnode *vnode) |
| 62 |
{ |
| 63 |
struct afs_io_locker *locker; |
| 64 |
|
| 65 |
spin_lock(&vnode->lock); |
| 66 |
locker = list_first_entry_or_null(&vnode->io_lock_waiters, |
| 67 |
struct afs_io_locker, link); |
| 68 |
if (locker) { |
| 69 |
list_del(&locker->link); |
| 70 |
smp_store_release(&locker->have_lock, 1); /* The unlock barrier. */ |
| 71 |
smp_mb__after_atomic(); /* Store have_lock before task state */ |
| 72 |
wake_up_process(locker->task); |
| 73 |
} else { |
| 74 |
clear_bit(AFS_VNODE_IO_LOCK, &vnode->flags); |
| 75 |
} |
| 76 |
spin_unlock(&vnode->lock); |
| 77 |
} |
| 78 |
|
| 79 |
/* |
| 80 |
* Lock the I/O lock on a vnode uninterruptibly. We can't use an ordinary |
| 81 |
* mutex as lockdep will complain if we unlock it in the wrong thread. |
| 82 |
*/ |
| 83 |
static void afs_lock_for_io(struct afs_vnode *vnode) |
| 84 |
{ |
| 85 |
struct afs_io_locker myself = { .task = current, }; |
| 86 |
|
| 87 |
spin_lock(&vnode->lock); |
| 88 |
|
| 89 |
if (!test_and_set_bit(AFS_VNODE_IO_LOCK, &vnode->flags)) { |
| 90 |
spin_unlock(&vnode->lock); |
| 91 |
return; |
| 92 |
} |
| 93 |
|
| 94 |
list_add_tail(&myself.link, &vnode->io_lock_waiters); |
| 95 |
spin_unlock(&vnode->lock); |
| 96 |
|
| 97 |
for (;;) { |
| 98 |
set_current_state(TASK_UNINTERRUPTIBLE); |
| 99 |
if (smp_load_acquire(&myself.have_lock)) /* The lock barrier */ |
| 100 |
break; |
| 101 |
schedule(); |
| 102 |
} |
| 103 |
__set_current_state(TASK_RUNNING); |
| 104 |
} |
| 105 |
|
| 106 |
/* |
| 107 |
* Lock the I/O lock on a vnode interruptibly. We can't use an ordinary mutex |
| 108 |
* as lockdep will complain if we unlock it in the wrong thread. |
| 109 |
*/ |
| 110 |
static int afs_lock_for_io_interruptible(struct afs_vnode *vnode) |
| 111 |
{ |
| 112 |
struct afs_io_locker myself = { .task = current, }; |
| 113 |
int ret = 0; |
| 114 |
|
| 115 |
spin_lock(&vnode->lock); |
| 116 |
|
| 117 |
if (!test_and_set_bit(AFS_VNODE_IO_LOCK, &vnode->flags)) { |
| 118 |
spin_unlock(&vnode->lock); |
| 119 |
return 0; |
| 120 |
} |
| 121 |
|
| 122 |
list_add_tail(&myself.link, &vnode->io_lock_waiters); |
| 123 |
spin_unlock(&vnode->lock); |
| 124 |
|
| 125 |
for (;;) { |
| 126 |
set_current_state(TASK_INTERRUPTIBLE); |
| 127 |
if (smp_load_acquire(&myself.have_lock) || /* The lock barrier */ |
| 128 |
signal_pending(current)) |
| 129 |
break; |
| 130 |
schedule(); |
| 131 |
} |
| 132 |
__set_current_state(TASK_RUNNING); |
| 133 |
|
| 134 |
/* If we got a signal, try to transfer the lock onto the next |
| 135 |
* waiter. |
| 136 |
*/ |
| 137 |
if (unlikely(signal_pending(current))) { |
| 138 |
spin_lock(&vnode->lock); |
| 139 |
if (myself.have_lock) { |
| 140 |
spin_unlock(&vnode->lock); |
| 141 |
afs_unlock_for_io(vnode); |
| 142 |
} else { |
| 143 |
list_del(&myself.link); |
| 144 |
spin_unlock(&vnode->lock); |
| 145 |
} |
| 146 |
ret = -ERESTARTSYS; |
| 147 |
} |
| 148 |
return ret; |
| 149 |
} |
| 150 |
|
| 151 |
/* |
| 152 |
* Lock the vnode(s) being operated upon. |
| 153 |
*/ |
| 154 |
static bool afs_get_io_locks(struct afs_operation *op) |
| 155 |
{ |
| 156 |
struct afs_vnode *vnode = op->file[0].vnode; |
| 157 |
struct afs_vnode *vnode2 = op->file[1].vnode; |
| 158 |
|
| 159 |
_enter(""); |
| 160 |
|
| 161 |
if (op->flags & AFS_OPERATION_UNINTR) { |
| 162 |
afs_lock_for_io(vnode); |
| 163 |
op->flags |= AFS_OPERATION_LOCK_0; |
| 164 |
_leave(" = t [1]"); |
| 165 |
return true; |
| 166 |
} |
| 167 |
|
| 168 |
if (!vnode2 || !op->file[1].need_io_lock || vnode == vnode2) |
| 169 |
vnode2 = NULL; |
| 170 |
|
| 171 |
if (vnode2 > vnode) |
| 172 |
swap(vnode, vnode2); |
| 173 |
|
| 174 |
if (afs_lock_for_io_interruptible(vnode) < 0) { |
| 175 |
afs_op_set_error(op, -ERESTARTSYS); |
| 176 |
op->flags |= AFS_OPERATION_STOP; |
| 177 |
_leave(" = f [I 0]"); |
| 178 |
return false; |
| 179 |
} |
| 180 |
op->flags |= AFS_OPERATION_LOCK_0; |
| 181 |
|
| 182 |
if (vnode2) { |
| 183 |
if (afs_lock_for_io_interruptible(vnode2) < 0) { |
| 184 |
afs_op_set_error(op, -ERESTARTSYS); |
| 185 |
op->flags |= AFS_OPERATION_STOP; |
| 186 |
afs_unlock_for_io(vnode); |
| 187 |
op->flags &= ~AFS_OPERATION_LOCK_0; |
| 188 |
_leave(" = f [I 1]"); |
| 189 |
return false; |
| 190 |
} |
| 191 |
op->flags |= AFS_OPERATION_LOCK_1; |
| 192 |
} |
| 193 |
|
| 194 |
_leave(" = t [2]"); |
| 195 |
return true; |
| 196 |
} |
| 197 |
|
| 198 |
static void afs_drop_io_locks(struct afs_operation *op) |
| 199 |
{ |
| 200 |
struct afs_vnode *vnode = op->file[0].vnode; |
| 201 |
struct afs_vnode *vnode2 = op->file[1].vnode; |
| 202 |
|
| 203 |
_enter(""); |
| 204 |
|
| 205 |
if (op->flags & AFS_OPERATION_LOCK_1) |
| 206 |
afs_unlock_for_io(vnode2); |
| 207 |
if (op->flags & AFS_OPERATION_LOCK_0) |
| 208 |
afs_unlock_for_io(vnode); |
| 209 |
} |
| 210 |
|
| 211 |
static void afs_prepare_vnode(struct afs_operation *op, struct afs_vnode_param *vp, |
| 212 |
unsigned int index) |
| 213 |
{ |
| 214 |
struct afs_vnode *vnode = vp->vnode; |
| 215 |
|
| 216 |
if (vnode) { |
| 217 |
vp->fid = vnode->fid; |
| 218 |
vp->dv_before = vnode->status.data_version; |
| 219 |
vp->cb_break_before = afs_calc_vnode_cb_break(vnode); |
| 220 |
if (vnode->lock_state != AFS_VNODE_LOCK_NONE) |
| 221 |
op->flags |= AFS_OPERATION_CUR_ONLY; |
| 222 |
if (vp->modification) |
| 223 |
set_bit(AFS_VNODE_MODIFYING, &vnode->flags); |
| 224 |
} |
| 225 |
|
| 226 |
if (vp->fid.vnode) |
| 227 |
_debug("PREP[%u] {%llx:%llu.%u}", |
| 228 |
index, vp->fid.vid, vp->fid.vnode, vp->fid.unique); |
| 229 |
} |
| 230 |
|
| 231 |
/* |
| 232 |
* Begin an operation on the fileserver. |
| 233 |
* |
| 234 |
* Fileserver operations are serialised on the server by vnode, so we serialise |
| 235 |
* them here also using the io_lock. |
| 236 |
*/ |
| 237 |
bool afs_begin_vnode_operation(struct afs_operation *op) |
| 238 |
{ |
| 239 |
struct afs_vnode *vnode = op->file[0].vnode; |
| 240 |
|
| 241 |
ASSERT(vnode); |
| 242 |
|
| 243 |
_enter(""); |
| 244 |
|
| 245 |
if (op->file[0].need_io_lock) |
| 246 |
if (!afs_get_io_locks(op)) |
| 247 |
return false; |
| 248 |
|
| 249 |
afs_prepare_vnode(op, &op->file[0], 0); |
| 250 |
afs_prepare_vnode(op, &op->file[1], 1); |
| 251 |
op->cb_v_break = atomic_read(&op->volume->cb_v_break); |
| 252 |
_leave(" = true"); |
| 253 |
return true; |
| 254 |
} |
| 255 |
|
| 256 |
/* |
| 257 |
* Tidy up a filesystem cursor and unlock the vnode. |
| 258 |
*/ |
| 259 |
void afs_end_vnode_operation(struct afs_operation *op) |
| 260 |
{ |
| 261 |
_enter(""); |
| 262 |
|
| 263 |
switch (afs_op_error(op)) { |
| 264 |
case -EDESTADDRREQ: |
| 265 |
case -EADDRNOTAVAIL: |
| 266 |
case -ENETUNREACH: |
| 267 |
case -EHOSTUNREACH: |
| 268 |
afs_dump_edestaddrreq(op); |
| 269 |
break; |
| 270 |
} |
| 271 |
|
| 272 |
afs_drop_io_locks(op); |
| 273 |
} |
| 274 |
|
| 275 |
/* |
| 276 |
* Wait for an in-progress operation to complete. |
| 277 |
*/ |
| 278 |
void afs_wait_for_operation(struct afs_operation *op) |
| 279 |
{ |
| 280 |
_enter(""); |
| 281 |
|
| 282 |
while (afs_select_fileserver(op)) { |
| 283 |
op->call_responded = false; |
| 284 |
op->call_error = 0; |
| 285 |
op->call_abort_code = 0; |
| 286 |
if (test_bit(AFS_SERVER_FL_IS_YFS, &op->server->flags) && |
| 287 |
op->ops->issue_yfs_rpc) |
| 288 |
op->ops->issue_yfs_rpc(op); |
| 289 |
else if (op->ops->issue_afs_rpc) |
| 290 |
op->ops->issue_afs_rpc(op); |
| 291 |
else |
| 292 |
op->call_error = -ENOTSUPP; |
| 293 |
|
| 294 |
if (op->call) { |
| 295 |
afs_wait_for_call_to_complete(op->call); |
| 296 |
op->call_abort_code = op->call->abort_code; |
| 297 |
op->call_error = op->call->error; |
| 298 |
op->call_responded = op->call->responded; |
| 299 |
afs_put_call(op->call); |
| 300 |
} |
| 301 |
} |
| 302 |
|
| 303 |
if (op->call_responded && op->server) |
| 304 |
set_bit(AFS_SERVER_FL_RESPONDING, &op->server->flags); |
| 305 |
|
| 306 |
if (!afs_op_error(op)) { |
| 307 |
_debug("success"); |
| 308 |
op->ops->success(op); |
| 309 |
} else if (op->cumul_error.aborted) { |
| 310 |
if (op->ops->aborted) |
| 311 |
op->ops->aborted(op); |
| 312 |
} else { |
| 313 |
if (op->ops->failed) |
| 314 |
op->ops->failed(op); |
| 315 |
} |
| 316 |
|
| 317 |
afs_end_vnode_operation(op); |
| 318 |
|
| 319 |
if (!afs_op_error(op) && op->ops->edit_dir) { |
| 320 |
_debug("edit_dir"); |
| 321 |
op->ops->edit_dir(op); |
| 322 |
} |
| 323 |
_leave(""); |
| 324 |
} |
| 325 |
|
| 326 |
/* |
| 327 |
* Dispose of an operation. |
| 328 |
*/ |
| 329 |
int afs_put_operation(struct afs_operation *op) |
| 330 |
{ |
| 331 |
struct afs_addr_list *alist; |
| 332 |
int i, ret = afs_op_error(op); |
| 333 |
|
| 334 |
_enter("op=%08x,%d", op->debug_id, ret); |
| 335 |
|
| 336 |
if (op->ops && op->ops->put) |
| 337 |
op->ops->put(op); |
| 338 |
if (op->file[0].modification) |
| 339 |
clear_bit(AFS_VNODE_MODIFYING, &op->file[0].vnode->flags); |
| 340 |
if (op->file[1].modification && op->file[1].vnode != op->file[0].vnode) |
| 341 |
clear_bit(AFS_VNODE_MODIFYING, &op->file[1].vnode->flags); |
| 342 |
if (op->file[0].put_vnode) |
| 343 |
iput(&op->file[0].vnode->netfs.inode); |
| 344 |
if (op->file[1].put_vnode) |
| 345 |
iput(&op->file[1].vnode->netfs.inode); |
| 346 |
|
| 347 |
if (op->more_files) { |
| 348 |
for (i = 0; i < op->nr_files - 2; i++) |
| 349 |
if (op->more_files[i].put_vnode) |
| 350 |
iput(&op->more_files[i].vnode->netfs.inode); |
| 351 |
kfree(op->more_files); |
| 352 |
} |
| 353 |
|
| 354 |
if (op->estate) { |
| 355 |
alist = op->estate->addresses; |
| 356 |
if (alist) { |
| 357 |
if (op->call_responded && |
| 358 |
op->addr_index != alist->preferred && |
| 359 |
test_bit(alist->preferred, &op->addr_tried)) |
| 360 |
WRITE_ONCE(alist->preferred, op->addr_index); |
| 361 |
} |
| 362 |
} |
| 363 |
|
| 364 |
afs_clear_server_states(op); |
| 365 |
afs_put_serverlist(op->net, op->server_list); |
| 366 |
afs_put_volume(op->volume, afs_volume_trace_put_put_op); |
| 367 |
key_put(op->key); |
| 368 |
kfree(op); |
| 369 |
return ret; |
| 370 |
} |
| 371 |
|
| 372 |
int afs_do_sync_operation(struct afs_operation *op) |
| 373 |
{ |
| 374 |
afs_begin_vnode_operation(op); |
| 375 |
afs_wait_for_operation(op); |
| 376 |
return afs_put_operation(op); |
| 377 |
} |