| 1 |
// SPDX-License-Identifier: GPL-2.0-or-later |
| 2 |
/* vnode and volume validity verification. |
| 3 |
* |
| 4 |
* Copyright (C) 2023 Red Hat, Inc. All Rights Reserved. |
| 5 |
* Written by David Howells ([email protected]) |
| 6 |
*/ |
| 7 |
|
| 8 |
#include <linux/kernel.h> |
| 9 |
#include <linux/module.h> |
| 10 |
#include <linux/sched.h> |
| 11 |
#include "internal.h" |
| 12 |
|
| 13 |
/* |
| 14 |
* Data validation is managed through a number of mechanisms from the server: |
| 15 |
* |
| 16 |
* (1) On first contact with a server (such as if it has just been rebooted), |
| 17 |
* the server sends us a CB.InitCallBackState* request. |
| 18 |
* |
| 19 |
* (2) On a RW volume, in response to certain vnode (inode)-accessing RPC |
| 20 |
* calls, the server maintains a time-limited per-vnode promise that it |
| 21 |
* will send us a CB.CallBack request if a third party alters the vnodes |
| 22 |
* accessed. |
| 23 |
* |
| 24 |
* Note that a vnode-level callbacks may also be sent for other reasons, |
| 25 |
* such as filelock release. |
| 26 |
* |
| 27 |
* (3) On a RO (or Backup) volume, in response to certain vnode-accessing RPC |
| 28 |
* calls, each server maintains a time-limited per-volume promise that it |
| 29 |
* will send us a CB.CallBack request if the RO volume is updated to a |
| 30 |
* snapshot of the RW volume ("vos release"). This is an atomic event |
| 31 |
* that cuts over all instances of the RO volume across multiple servers |
| 32 |
* simultaneously. |
| 33 |
* |
| 34 |
* Note that a volume-level callbacks may also be sent for other reasons, |
| 35 |
* such as the volumeserver taking over control of the volume from the |
| 36 |
* fileserver. |
| 37 |
* |
| 38 |
* Note also that each server maintains an independent time limit on an |
| 39 |
* independent callback. |
| 40 |
* |
| 41 |
* (4) Certain RPC calls include a volume information record "VolSync" in |
| 42 |
* their reply. This contains a creation date for the volume that should |
| 43 |
* remain unchanged for a RW volume (but will be changed if the volume is |
| 44 |
* restored from backup) or will be bumped to the time of snapshotting |
| 45 |
* when a RO volume is released. |
| 46 |
* |
| 47 |
* In order to track this events, the following are provided: |
| 48 |
* |
| 49 |
* ->cb_v_break. A counter of events that might mean that the contents of |
| 50 |
* a volume have been altered since we last checked a vnode. |
| 51 |
* |
| 52 |
* ->cb_v_check. A counter of the number of events that we've sent a |
| 53 |
* query to the server for. Everything's up to date if this equals |
| 54 |
* cb_v_break. |
| 55 |
* |
| 56 |
* ->cb_scrub. A counter of the number of regression events for which we |
| 57 |
* have to completely wipe the cache. |
| 58 |
* |
| 59 |
* ->cb_ro_snapshot. A counter of the number of times that we've |
| 60 |
* recognised that a RO volume has been updated. |
| 61 |
* |
| 62 |
* ->cb_break. A counter of events that might mean that the contents of a |
| 63 |
* vnode have been altered. |
| 64 |
* |
| 65 |
* ->cb_expires_at. The time at which the callback promise expires or |
| 66 |
* AFS_NO_CB_PROMISE if we have no promise. |
| 67 |
* |
| 68 |
* The way we manage things is: |
| 69 |
* |
| 70 |
* (1) When a volume-level CB.CallBack occurs, we increment ->cb_v_break on |
| 71 |
* the volume and reset ->cb_expires_at (ie. set AFS_NO_CB_PROMISE) on the |
| 72 |
* volume and volume's server record. |
| 73 |
* |
| 74 |
* (2) When a CB.InitCallBackState occurs, we treat this as a volume-level |
| 75 |
* callback break on all the volumes that have been using that volume |
| 76 |
* (ie. increment ->cb_v_break and reset ->cb_expires_at). |
| 77 |
* |
| 78 |
* (3) When a vnode-level CB.CallBack occurs, we increment ->cb_break on the |
| 79 |
* vnode and reset its ->cb_expires_at. If the vnode is mmapped, we also |
| 80 |
* dispatch a work item to unmap all PTEs to the vnode's pagecache to |
| 81 |
* force reentry to the filesystem for revalidation. |
| 82 |
* |
| 83 |
* (4) When entering the filesystem, we call afs_validate() to check the |
| 84 |
* validity of a vnode. This first checks to see if ->cb_v_check and |
| 85 |
* ->cb_v_break match, and if they don't, we lock volume->cb_check_lock |
| 86 |
* exclusively and perform an FS.FetchStatus on the vnode. |
| 87 |
* |
| 88 |
* After checking the volume, we check the vnode. If there's a mismatch |
| 89 |
* between the volume counters and the vnode's mirrors of those counters, |
| 90 |
* we lock vnode->validate_lock and issue an FS.FetchStatus on the vnode. |
| 91 |
* |
| 92 |
* (5) When the reply from FS.FetchStatus arrives, the VolSync record is |
| 93 |
* parsed: |
| 94 |
* |
| 95 |
* (A) If the Creation timestamp has changed on a RW volume or regressed |
| 96 |
* on a RO volume, we try to increment ->cb_scrub; if it advances on a |
| 97 |
* RO volume, we assume "vos release" happened and try to increment |
| 98 |
* ->cb_ro_snapshot. |
| 99 |
* |
| 100 |
* (B) If the Update timestamp has regressed, we try to increment |
| 101 |
* ->cb_scrub. |
| 102 |
* |
| 103 |
* Note that in both of these cases, we only do the increment if we can |
| 104 |
* cmpxchg the value of the timestamp from the value we noted before the |
| 105 |
* op. This tries to prevent parallel ops from fighting one another. |
| 106 |
* |
| 107 |
* volume->cb_v_check is then set to ->cb_v_break. |
| 108 |
* |
| 109 |
* (6) The AFSCallBack record included in the FS.FetchStatus reply is also |
| 110 |
* parsed and used to set the promise in ->cb_expires_at for the vnode, |
| 111 |
* the volume and the volume's server record. |
| 112 |
* |
| 113 |
* (7) If ->cb_scrub is seen to have advanced, we invalidate the pagecache for |
| 114 |
* the vnode. |
| 115 |
*/ |
| 116 |
|
| 117 |
/* |
| 118 |
* Check the validity of a vnode/inode and its parent volume. |
| 119 |
*/ |
| 120 |
bool afs_check_validity(const struct afs_vnode *vnode) |
| 121 |
{ |
| 122 |
const struct afs_volume *volume = vnode->volume; |
| 123 |
enum afs_vnode_invalid_trace trace = afs_vnode_valid_trace; |
| 124 |
time64_t cb_expires_at = atomic64_read(&vnode->cb_expires_at); |
| 125 |
time64_t deadline = ktime_get_real_seconds() + 10; |
| 126 |
|
| 127 |
if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) |
| 128 |
return true; |
| 129 |
|
| 130 |
if (atomic_read(&volume->cb_v_check) != atomic_read(&volume->cb_v_break)) |
| 131 |
trace = afs_vnode_invalid_trace_cb_v_break; |
| 132 |
else if (cb_expires_at == AFS_NO_CB_PROMISE) |
| 133 |
trace = afs_vnode_invalid_trace_no_cb_promise; |
| 134 |
else if (cb_expires_at <= deadline) |
| 135 |
trace = afs_vnode_invalid_trace_expired; |
| 136 |
else if (volume->cb_expires_at <= deadline) |
| 137 |
trace = afs_vnode_invalid_trace_vol_expired; |
| 138 |
else if (vnode->cb_ro_snapshot != atomic_read(&volume->cb_ro_snapshot)) |
| 139 |
trace = afs_vnode_invalid_trace_cb_ro_snapshot; |
| 140 |
else if (vnode->cb_scrub != atomic_read(&volume->cb_scrub)) |
| 141 |
trace = afs_vnode_invalid_trace_cb_scrub; |
| 142 |
else if (test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) |
| 143 |
trace = afs_vnode_invalid_trace_zap_data; |
| 144 |
else |
| 145 |
return true; |
| 146 |
trace_afs_vnode_invalid(vnode, trace); |
| 147 |
return false; |
| 148 |
} |
| 149 |
|
| 150 |
/* |
| 151 |
* See if the server we've just talked to is currently excluded. |
| 152 |
*/ |
| 153 |
static bool __afs_is_server_excluded(struct afs_operation *op, struct afs_volume *volume) |
| 154 |
{ |
| 155 |
const struct afs_server_entry *se; |
| 156 |
const struct afs_server_list *slist; |
| 157 |
bool is_excluded = true; |
| 158 |
int i; |
| 159 |
|
| 160 |
rcu_read_lock(); |
| 161 |
|
| 162 |
slist = rcu_dereference(volume->servers); |
| 163 |
for (i = 0; i < slist->nr_servers; i++) { |
| 164 |
se = &slist->servers[i]; |
| 165 |
if (op->server == se->server) { |
| 166 |
is_excluded = test_bit(AFS_SE_EXCLUDED, &se->flags); |
| 167 |
break; |
| 168 |
} |
| 169 |
} |
| 170 |
|
| 171 |
rcu_read_unlock(); |
| 172 |
return is_excluded; |
| 173 |
} |
| 174 |
|
| 175 |
/* |
| 176 |
* Update the volume's server list when the creation time changes and see if |
| 177 |
* the server we've just talked to is currently excluded. |
| 178 |
*/ |
| 179 |
static int afs_is_server_excluded(struct afs_operation *op, struct afs_volume *volume) |
| 180 |
{ |
| 181 |
int ret; |
| 182 |
|
| 183 |
if (__afs_is_server_excluded(op, volume)) |
| 184 |
return 1; |
| 185 |
|
| 186 |
set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags); |
| 187 |
ret = afs_check_volume_status(op->volume, op); |
| 188 |
if (ret < 0) |
| 189 |
return ret; |
| 190 |
|
| 191 |
return __afs_is_server_excluded(op, volume); |
| 192 |
} |
| 193 |
|
| 194 |
/* |
| 195 |
* Handle a change to the volume creation time in the VolSync record. |
| 196 |
*/ |
| 197 |
static int afs_update_volume_creation_time(struct afs_operation *op, struct afs_volume *volume) |
| 198 |
{ |
| 199 |
unsigned int snap; |
| 200 |
time64_t cur = volume->creation_time; |
| 201 |
time64_t old = op->pre_volsync.creation; |
| 202 |
time64_t new = op->volsync.creation; |
| 203 |
int ret; |
| 204 |
|
| 205 |
_enter("%llx,%llx,%llx->%llx", volume->vid, cur, old, new); |
| 206 |
|
| 207 |
if (cur == TIME64_MIN) { |
| 208 |
volume->creation_time = new; |
| 209 |
return 0; |
| 210 |
} |
| 211 |
|
| 212 |
if (new == cur) |
| 213 |
return 0; |
| 214 |
|
| 215 |
/* Try to advance the creation timestamp from what we had before the |
| 216 |
* operation to what we got back from the server. This should |
| 217 |
* hopefully ensure that in a race between multiple operations only one |
| 218 |
* of them will do this. |
| 219 |
*/ |
| 220 |
if (cur != old) |
| 221 |
return 0; |
| 222 |
|
| 223 |
/* If the creation time changes in an unexpected way, we need to scrub |
| 224 |
* our caches. For a RW vol, this will only change if the volume is |
| 225 |
* restored from a backup; for a RO/Backup vol, this will advance when |
| 226 |
* the volume is updated to a new snapshot (eg. "vos release"). |
| 227 |
*/ |
| 228 |
if (volume->type == AFSVL_RWVOL) |
| 229 |
goto regressed; |
| 230 |
if (volume->type == AFSVL_BACKVOL) { |
| 231 |
if (new < old) |
| 232 |
goto regressed; |
| 233 |
goto advance; |
| 234 |
} |
| 235 |
|
| 236 |
/* We have an RO volume, we need to query the VL server and look at the |
| 237 |
* server flags to see if RW->RO replication is in progress. |
| 238 |
*/ |
| 239 |
ret = afs_is_server_excluded(op, volume); |
| 240 |
if (ret < 0) |
| 241 |
return ret; |
| 242 |
if (ret > 0) { |
| 243 |
snap = atomic_read(&volume->cb_ro_snapshot); |
| 244 |
trace_afs_cb_v_break(volume->vid, snap, afs_cb_break_volume_excluded); |
| 245 |
return ret; |
| 246 |
} |
| 247 |
|
| 248 |
advance: |
| 249 |
snap = atomic_inc_return(&volume->cb_ro_snapshot); |
| 250 |
trace_afs_cb_v_break(volume->vid, snap, afs_cb_break_for_vos_release); |
| 251 |
volume->creation_time = new; |
| 252 |
return 0; |
| 253 |
|
| 254 |
regressed: |
| 255 |
atomic_inc(&volume->cb_scrub); |
| 256 |
trace_afs_cb_v_break(volume->vid, 0, afs_cb_break_for_creation_regress); |
| 257 |
volume->creation_time = new; |
| 258 |
return 0; |
| 259 |
} |
| 260 |
|
| 261 |
/* |
| 262 |
* Handle a change to the volume update time in the VolSync record. |
| 263 |
*/ |
| 264 |
static void afs_update_volume_update_time(struct afs_operation *op, struct afs_volume *volume) |
| 265 |
{ |
| 266 |
enum afs_cb_break_reason reason = afs_cb_break_no_break; |
| 267 |
time64_t cur = volume->update_time; |
| 268 |
time64_t old = op->pre_volsync.update; |
| 269 |
time64_t new = op->volsync.update; |
| 270 |
|
| 271 |
_enter("%llx,%llx,%llx->%llx", volume->vid, cur, old, new); |
| 272 |
|
| 273 |
if (cur == TIME64_MIN) { |
| 274 |
volume->update_time = new; |
| 275 |
return; |
| 276 |
} |
| 277 |
|
| 278 |
if (new == cur) |
| 279 |
return; |
| 280 |
|
| 281 |
/* If the volume update time changes in an unexpected way, we need to |
| 282 |
* scrub our caches. For a RW vol, this will advance on every |
| 283 |
* modification op; for a RO/Backup vol, this will advance when the |
| 284 |
* volume is updated to a new snapshot (eg. "vos release"). |
| 285 |
*/ |
| 286 |
if (new < old) |
| 287 |
reason = afs_cb_break_for_update_regress; |
| 288 |
|
| 289 |
/* Try to advance the update timestamp from what we had before the |
| 290 |
* operation to what we got back from the server. This should |
| 291 |
* hopefully ensure that in a race between multiple operations only one |
| 292 |
* of them will do this. |
| 293 |
*/ |
| 294 |
if (cur == old) { |
| 295 |
if (reason == afs_cb_break_for_update_regress) { |
| 296 |
atomic_inc(&volume->cb_scrub); |
| 297 |
trace_afs_cb_v_break(volume->vid, 0, reason); |
| 298 |
} |
| 299 |
volume->update_time = new; |
| 300 |
} |
| 301 |
} |
| 302 |
|
| 303 |
static int afs_update_volume_times(struct afs_operation *op, struct afs_volume *volume) |
| 304 |
{ |
| 305 |
int ret = 0; |
| 306 |
|
| 307 |
if (likely(op->volsync.creation == volume->creation_time && |
| 308 |
op->volsync.update == volume->update_time)) |
| 309 |
return 0; |
| 310 |
|
| 311 |
mutex_lock(&volume->volsync_lock); |
| 312 |
if (op->volsync.creation != volume->creation_time) { |
| 313 |
ret = afs_update_volume_creation_time(op, volume); |
| 314 |
if (ret < 0) |
| 315 |
goto out; |
| 316 |
} |
| 317 |
if (op->volsync.update != volume->update_time) |
| 318 |
afs_update_volume_update_time(op, volume); |
| 319 |
out: |
| 320 |
mutex_unlock(&volume->volsync_lock); |
| 321 |
return ret; |
| 322 |
} |
| 323 |
|
| 324 |
/* |
| 325 |
* Update the state of a volume, including recording the expiration time of the |
| 326 |
* callback promise. Returns 1 to redo the operation from the start. |
| 327 |
*/ |
| 328 |
int afs_update_volume_state(struct afs_operation *op) |
| 329 |
{ |
| 330 |
struct afs_server_list *slist = op->server_list; |
| 331 |
struct afs_server_entry *se = &slist->servers[op->server_index]; |
| 332 |
struct afs_callback *cb = &op->file[0].scb.callback; |
| 333 |
struct afs_volume *volume = op->volume; |
| 334 |
unsigned int cb_v_break = atomic_read(&volume->cb_v_break); |
| 335 |
unsigned int cb_v_check = atomic_read(&volume->cb_v_check); |
| 336 |
int ret; |
| 337 |
|
| 338 |
_enter("%llx", op->volume->vid); |
| 339 |
|
| 340 |
if (op->volsync.creation != TIME64_MIN || op->volsync.update != TIME64_MIN) { |
| 341 |
ret = afs_update_volume_times(op, volume); |
| 342 |
if (ret != 0) { |
| 343 |
_leave(" = %d", ret); |
| 344 |
return ret; |
| 345 |
} |
| 346 |
} |
| 347 |
|
| 348 |
if (op->cb_v_break == cb_v_break && |
| 349 |
(op->file[0].scb.have_cb || op->file[1].scb.have_cb)) { |
| 350 |
time64_t expires_at = cb->expires_at; |
| 351 |
|
| 352 |
if (!op->file[0].scb.have_cb) |
| 353 |
expires_at = op->file[1].scb.callback.expires_at; |
| 354 |
|
| 355 |
se->cb_expires_at = expires_at; |
| 356 |
volume->cb_expires_at = expires_at; |
| 357 |
} |
| 358 |
if (cb_v_check < op->cb_v_break) |
| 359 |
atomic_cmpxchg(&volume->cb_v_check, cb_v_check, op->cb_v_break); |
| 360 |
return 0; |
| 361 |
} |
| 362 |
|
| 363 |
/* |
| 364 |
* mark the data attached to an inode as obsolete due to a write on the server |
| 365 |
* - might also want to ditch all the outstanding writes and dirty pages |
| 366 |
*/ |
| 367 |
static void afs_zap_data(struct afs_vnode *vnode) |
| 368 |
{ |
| 369 |
_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode); |
| 370 |
|
| 371 |
afs_invalidate_cache(vnode, 0); |
| 372 |
|
| 373 |
/* nuke all the non-dirty pages that aren't locked, mapped or being |
| 374 |
* written back in a regular file and completely discard the pages in a |
| 375 |
* directory or symlink */ |
| 376 |
if (S_ISREG(vnode->netfs.inode.i_mode)) |
| 377 |
filemap_invalidate_inode(&vnode->netfs.inode, true, 0, LLONG_MAX); |
| 378 |
else |
| 379 |
filemap_invalidate_inode(&vnode->netfs.inode, false, 0, LLONG_MAX); |
| 380 |
} |
| 381 |
|
| 382 |
/* |
| 383 |
* validate a vnode/inode |
| 384 |
* - there are several things we need to check |
| 385 |
* - parent dir data changes (rm, rmdir, rename, mkdir, create, link, |
| 386 |
* symlink) |
| 387 |
* - parent dir metadata changed (security changes) |
| 388 |
* - dentry data changed (write, truncate) |
| 389 |
* - dentry metadata changed (security changes) |
| 390 |
*/ |
| 391 |
int afs_validate(struct afs_vnode *vnode, struct key *key) |
| 392 |
{ |
| 393 |
struct afs_volume *volume = vnode->volume; |
| 394 |
unsigned int cb_ro_snapshot, cb_scrub; |
| 395 |
time64_t deadline = ktime_get_real_seconds() + 10; |
| 396 |
bool zap = false, locked_vol = false; |
| 397 |
int ret; |
| 398 |
|
| 399 |
_enter("{v={%llx:%llu} fl=%lx},%x", |
| 400 |
vnode->fid.vid, vnode->fid.vnode, vnode->flags, |
| 401 |
key_serial(key)); |
| 402 |
|
| 403 |
if (afs_check_validity(vnode)) |
| 404 |
return test_bit(AFS_VNODE_DELETED, &vnode->flags) ? -ESTALE : 0; |
| 405 |
|
| 406 |
ret = down_write_killable(&vnode->validate_lock); |
| 407 |
if (ret < 0) |
| 408 |
goto error; |
| 409 |
|
| 410 |
if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) { |
| 411 |
ret = -ESTALE; |
| 412 |
goto error_unlock; |
| 413 |
} |
| 414 |
|
| 415 |
/* Validate a volume after the v_break has changed or the volume |
| 416 |
* callback expired. We only want to do this once per volume per |
| 417 |
* v_break change. The actual work will be done when parsing the |
| 418 |
* status fetch reply. |
| 419 |
*/ |
| 420 |
if (volume->cb_expires_at <= deadline || |
| 421 |
atomic_read(&volume->cb_v_check) != atomic_read(&volume->cb_v_break)) { |
| 422 |
ret = mutex_lock_interruptible(&volume->cb_check_lock); |
| 423 |
if (ret < 0) |
| 424 |
goto error_unlock; |
| 425 |
locked_vol = true; |
| 426 |
} |
| 427 |
|
| 428 |
cb_ro_snapshot = atomic_read(&volume->cb_ro_snapshot); |
| 429 |
cb_scrub = atomic_read(&volume->cb_scrub); |
| 430 |
if (vnode->cb_ro_snapshot != cb_ro_snapshot || |
| 431 |
vnode->cb_scrub != cb_scrub) |
| 432 |
unmap_mapping_pages(vnode->netfs.inode.i_mapping, 0, 0, false); |
| 433 |
|
| 434 |
if (vnode->cb_ro_snapshot != cb_ro_snapshot || |
| 435 |
vnode->cb_scrub != cb_scrub || |
| 436 |
volume->cb_expires_at <= deadline || |
| 437 |
atomic_read(&volume->cb_v_check) != atomic_read(&volume->cb_v_break) || |
| 438 |
atomic64_read(&vnode->cb_expires_at) <= deadline |
| 439 |
) { |
| 440 |
ret = afs_fetch_status(vnode, key, false, NULL); |
| 441 |
if (ret < 0) { |
| 442 |
if (ret == -ENOENT) { |
| 443 |
set_bit(AFS_VNODE_DELETED, &vnode->flags); |
| 444 |
ret = -ESTALE; |
| 445 |
} |
| 446 |
goto error_unlock; |
| 447 |
} |
| 448 |
|
| 449 |
_debug("new promise [fl=%lx]", vnode->flags); |
| 450 |
} |
| 451 |
|
| 452 |
/* We can drop the volume lock now as. */ |
| 453 |
if (locked_vol) { |
| 454 |
mutex_unlock(&volume->cb_check_lock); |
| 455 |
locked_vol = false; |
| 456 |
} |
| 457 |
|
| 458 |
cb_ro_snapshot = atomic_read(&volume->cb_ro_snapshot); |
| 459 |
cb_scrub = atomic_read(&volume->cb_scrub); |
| 460 |
_debug("vnode inval %x==%x %x==%x", |
| 461 |
vnode->cb_ro_snapshot, cb_ro_snapshot, |
| 462 |
vnode->cb_scrub, cb_scrub); |
| 463 |
if (vnode->cb_scrub != cb_scrub) |
| 464 |
zap = true; |
| 465 |
vnode->cb_ro_snapshot = cb_ro_snapshot; |
| 466 |
vnode->cb_scrub = cb_scrub; |
| 467 |
|
| 468 |
/* if the vnode's data version number changed then its contents are |
| 469 |
* different */ |
| 470 |
zap |= test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags); |
| 471 |
if (zap) |
| 472 |
afs_zap_data(vnode); |
| 473 |
up_write(&vnode->validate_lock); |
| 474 |
_leave(" = 0"); |
| 475 |
return 0; |
| 476 |
|
| 477 |
error_unlock: |
| 478 |
if (locked_vol) |
| 479 |
mutex_unlock(&volume->cb_check_lock); |
| 480 |
up_write(&vnode->validate_lock); |
| 481 |
error: |
| 482 |
_leave(" = %d", ret); |
| 483 |
return ret; |
| 484 |
} |