| 1 |
/* -*- linux-c -*- |
| 2 |
* |
| 3 |
* This transport version uses relayfs on top of a debugfs or procfs |
| 4 |
* file. This code started as a proposed relayfs interface called |
| 5 |
* 'utt'. It has been modified and simplified for systemtap. |
| 6 |
* |
| 7 |
* Changes Copyright (C) 2009-2020 Red Hat Inc. |
| 8 |
* |
| 9 |
* Original utt code by: |
| 10 |
* Copyright (C) 2006 Jens Axboe <[email protected]> |
| 11 |
* Moved to utt.c by Tom Zanussi, 2006 |
| 12 |
* |
| 13 |
* This program is free software; you can redistribute it and/or modify |
| 14 |
* it under the terms of the GNU General Public License version 2 as |
| 15 |
* published by the Free Software Foundation. |
| 16 |
* |
| 17 |
* This program is distributed in the hope that it will be useful, |
| 18 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 19 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 20 |
* GNU General Public License for more details. |
| 21 |
* |
| 22 |
* You should have received a copy of the GNU General Public License |
| 23 |
* along with this program. If not, see <http://www.gnu.org/licenses/>. |
| 24 |
* |
| 25 |
*/ |
| 26 |
#include <linux/kernel.h> |
| 27 |
#include <linux/module.h> |
| 28 |
#include <linux/percpu.h> |
| 29 |
#include <linux/init.h> |
| 30 |
#include <linux/debugfs.h> |
| 31 |
#include <linux/mm.h> |
| 32 |
#include "../linux/timer_compatibility.h" |
| 33 |
#include "../uidgid_compatibility.h" |
| 34 |
#include "relay_compat.h" |
| 35 |
|
| 36 |
#ifndef STP_RELAY_TIMER_INTERVAL |
| 37 |
/* Wakeup timer interval in jiffies (default 1 ms) */ |
| 38 |
#define STP_RELAY_TIMER_INTERVAL ((HZ + 999) / 1000) |
| 39 |
#endif |
| 40 |
|
| 41 |
/* Note: if struct _stp_relay_data_type changes, staplog.c might need |
| 42 |
* to be changed. */ |
| 43 |
struct _stp_relay_data_type { |
| 44 |
struct rchan *rchan; |
| 45 |
atomic_t /* enum _stp_transport_state */ transport_state; |
| 46 |
atomic_t wakeup; |
| 47 |
struct timer_list timer; |
| 48 |
int overwrite_flag; |
| 49 |
}; |
| 50 |
struct _stp_relay_data_type _stp_relay_data; |
| 51 |
|
| 52 |
/* relay_file_operations is const, so .owner is obviously not set there. |
| 53 |
* Below struct, filled in _stp_transport_data_fs_init(), fixes it. */ |
| 54 |
static struct file_operations relay_file_operations_w_owner; |
| 55 |
|
| 56 |
/* |
| 57 |
* __stp_relay_switch_subbuf - switch to a new sub-buffer |
| 58 |
* |
| 59 |
* Most of this function is deadcopy of relay_switch_subbuf. |
| 60 |
*/ |
| 61 |
static size_t __stp_relay_switch_subbuf(struct rchan_buf *buf, size_t length) |
| 62 |
{ |
| 63 |
char *old, *new; |
| 64 |
size_t old_subbuf, new_subbuf; |
| 65 |
#if !defined(STAPCONF_PR33324_PREV_PADDING) |
| 66 |
size_t prev_padding; |
| 67 |
#endif |
| 68 |
|
| 69 |
if (unlikely(buf == NULL)) |
| 70 |
return 0; |
| 71 |
|
| 72 |
if (unlikely(length > buf->chan->subbuf_size)) |
| 73 |
length = buf->chan->subbuf_size; |
| 74 |
|
| 75 |
if (buf->offset != buf->chan->subbuf_size + 1) { |
| 76 |
#if defined(STAPCONF_PR33324_PREV_PADDING) |
| 77 |
buf->prev_padding = buf->chan->subbuf_size - buf->offset; |
| 78 |
old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; |
| 79 |
buf->padding[old_subbuf] = buf->prev_padding; |
| 80 |
#else |
| 81 |
prev_padding = buf->chan->subbuf_size - buf->offset; |
| 82 |
old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; |
| 83 |
buf->padding[old_subbuf] = prev_padding; |
| 84 |
#endif |
| 85 |
buf->subbufs_produced++; |
| 86 |
buf->dentry->d_inode->i_size += buf->chan->subbuf_size - |
| 87 |
buf->padding[old_subbuf]; |
| 88 |
smp_mb(); |
| 89 |
} |
| 90 |
|
| 91 |
old = buf->data; |
| 92 |
new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; |
| 93 |
new = (char*)buf->start + new_subbuf * buf->chan->subbuf_size; |
| 94 |
buf->offset = 0; |
| 95 |
#if defined(STAPCONF_PR33324_PREV_PADDING) |
| 96 |
if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) { |
| 97 |
#else |
| 98 |
if (!buf->chan->cb->subbuf_start(buf, new, old)) { |
| 99 |
#endif |
| 100 |
buf->offset = buf->chan->subbuf_size + 1; |
| 101 |
return 0; |
| 102 |
} |
| 103 |
buf->data = new; |
| 104 |
buf->padding[new_subbuf] = 0; |
| 105 |
|
| 106 |
return length; |
| 107 |
} |
| 108 |
|
| 109 |
static bool __stp_relay_buf_empty(struct rchan_buf *buf) |
| 110 |
{ |
| 111 |
return !buf->offset && buf->subbufs_produced == buf->subbufs_consumed; |
| 112 |
} |
| 113 |
|
| 114 |
static int __stp_relay_file_open(struct inode *inode, struct file *filp) |
| 115 |
{ |
| 116 |
struct rchan_buf *buf = inode->i_private; |
| 117 |
|
| 118 |
/* kernel's relay_file_open() does not check NULL buf (at least as of |
| 119 |
* Sep 9, 2022) and there is a race window in our |
| 120 |
* __stp_procfs_relay_create_buf_file_callback() function, |
| 121 |
* between proc_create() and the `in->i_private = buf` assignment. |
| 122 |
* without this check, we can reproduce a NULL ptr deref kernel panic |
| 123 |
* while loading a system with a lot of short-lived stap sessions. |
| 124 |
*/ |
| 125 |
if (unlikely(buf == NULL)) { |
| 126 |
dbug_trans(0, "relay_file_open: found NULL rchan buf\n"); |
| 127 |
return -ENOENT; |
| 128 |
} |
| 129 |
|
| 130 |
return relay_file_operations.open(inode, filp); |
| 131 |
} |
| 132 |
|
| 133 |
static unsigned int __stp_relay_file_poll(struct file *filp, poll_table *wait) |
| 134 |
{ |
| 135 |
struct rchan_buf *buf = filp->private_data; |
| 136 |
|
| 137 |
if (buf->finalized) |
| 138 |
return POLLERR; |
| 139 |
|
| 140 |
if (filp->f_mode & FMODE_READ) { |
| 141 |
poll_wait(filp, &buf->read_wait, wait); |
| 142 |
if (!__stp_relay_buf_empty(buf)) |
| 143 |
return POLLIN | POLLRDNORM; |
| 144 |
} |
| 145 |
|
| 146 |
return 0; |
| 147 |
} |
| 148 |
|
| 149 |
static ssize_t __stp_relay_file_read(struct file *filp, char __user *buffer, |
| 150 |
size_t count, loff_t *ppos) |
| 151 |
{ |
| 152 |
struct rchan_buf *buf = filp->private_data; |
| 153 |
unsigned long flags; |
| 154 |
|
| 155 |
/* This trylock will only fail after the print driver is destroyed */ |
| 156 |
if (_stp_print_trylock_irqsave(&flags)) { |
| 157 |
/* Switch out the current buffer if it has any data */ |
| 158 |
if (buf->offset) |
| 159 |
__stp_relay_switch_subbuf(buf, 0); |
| 160 |
_stp_print_unlock_irqrestore(&flags); |
| 161 |
} |
| 162 |
|
| 163 |
/* |
| 164 |
* Optimization: bail out early if there's nothing to read. This is |
| 165 |
* faster than going into relay's read() function without having any |
| 166 |
* data, as it will acquire a mutex lock for the inode before bailing. |
| 167 |
*/ |
| 168 |
if (buf->subbufs_produced == buf->subbufs_consumed) |
| 169 |
return -ENODATA; |
| 170 |
|
| 171 |
return relay_file_operations.read(filp, buffer, count, ppos); |
| 172 |
} |
| 173 |
|
| 174 |
static void __stp_relay_wakeup_readers(struct rchan_buf *buf) |
| 175 |
{ |
| 176 |
if (buf && waitqueue_active(&buf->read_wait) && |
| 177 |
!__stp_relay_buf_empty(buf)) |
| 178 |
wake_up_interruptible(&buf->read_wait); |
| 179 |
} |
| 180 |
|
| 181 |
static void __stp_relay_wakeup_timer(stp_timer_callback_parameter_t unused) |
| 182 |
{ |
| 183 |
int i; |
| 184 |
|
| 185 |
if (atomic_cmpxchg(&_stp_relay_data.wakeup, 1, 0)) { |
| 186 |
struct rchan_buf *buf; |
| 187 |
|
| 188 |
/* NB it makes no sense to wake up readers on offline CPUs */ |
| 189 |
for_each_online_cpu(i) { |
| 190 |
buf = _stp_get_rchan_subbuf(_stp_relay_data.rchan->buf, |
| 191 |
i); |
| 192 |
|
| 193 |
/* relay_open() only initializes bufs on the online CPUs |
| 194 |
* at the time of invocation. The online CPUs may |
| 195 |
* change at any time, so the current online CPU might |
| 196 |
* be offline when relay_open() was called. so we must |
| 197 |
* check if the buf for the current CPU is invalid |
| 198 |
* otherwise we may dereference a NULL pointer. |
| 199 |
*/ |
| 200 |
if (likely(buf)) |
| 201 |
__stp_relay_wakeup_readers(buf); |
| 202 |
} |
| 203 |
} |
| 204 |
|
| 205 |
if (atomic_read(&_stp_relay_data.transport_state) == STP_TRANSPORT_RUNNING) |
| 206 |
mod_timer(&_stp_relay_data.timer, jiffies + STP_RELAY_TIMER_INTERVAL); |
| 207 |
else |
| 208 |
dbug_trans(0, "relay_v2 wakeup timer expiry\n"); |
| 209 |
} |
| 210 |
|
| 211 |
static void __stp_relay_timer_init(void) |
| 212 |
{ |
| 213 |
atomic_set(&_stp_relay_data.wakeup, 0); |
| 214 |
timer_setup(&_stp_relay_data.timer, __stp_relay_wakeup_timer, 0); |
| 215 |
_stp_relay_data.timer.expires = jiffies + STP_RELAY_TIMER_INTERVAL; |
| 216 |
add_timer(&_stp_relay_data.timer); |
| 217 |
smp_mb(); |
| 218 |
} |
| 219 |
|
| 220 |
static enum _stp_transport_state _stp_transport_get_state(void) |
| 221 |
{ |
| 222 |
return atomic_read (&_stp_relay_data.transport_state); |
| 223 |
} |
| 224 |
|
| 225 |
static void _stp_transport_data_fs_overwrite(int overwrite) |
| 226 |
{ |
| 227 |
_stp_relay_data.overwrite_flag = overwrite; |
| 228 |
} |
| 229 |
|
| 230 |
|
| 231 |
/* |
| 232 |
* Keep track of how many times we encountered a full subbuffer, to aid |
| 233 |
* the user space app in telling how many lost events there were. |
| 234 |
*/ |
| 235 |
#if defined(STAPCONF_PR33324_PREV_PADDING) |
| 236 |
static int __stp_relay_subbuf_start_callback(struct rchan_buf *buf, |
| 237 |
void *subbuf, void *prev_subbuf, |
| 238 |
size_t prev_padding) |
| 239 |
#else |
| 240 |
static int __stp_relay_subbuf_start_callback(struct rchan_buf *buf, |
| 241 |
void *subbuf, void *prev_subbuf) |
| 242 |
#endif |
| 243 |
{ |
| 244 |
if (_stp_relay_data.overwrite_flag || !relay_buf_full(buf)) |
| 245 |
return 1; |
| 246 |
|
| 247 |
#ifdef _STP_USE_DROPPED_FILE |
| 248 |
atomic_inc(&_stp_relay_data.dropped); |
| 249 |
#endif |
| 250 |
return 0; |
| 251 |
} |
| 252 |
|
| 253 |
|
| 254 |
// PR26665: demultiplex debugfs vs procfs host |
| 255 |
|
| 256 |
static int __stp_relay_remove_buf_file_callback(struct dentry *dentry) |
| 257 |
{ |
| 258 |
if (debugfs_p) |
| 259 |
return __stp_debugfs_relay_remove_buf_file_callback(dentry); |
| 260 |
if (procfs_p) |
| 261 |
return __stp_procfs_relay_remove_buf_file_callback(dentry); |
| 262 |
return 0; |
| 263 |
} |
| 264 |
|
| 265 |
|
| 266 |
static struct dentry * |
| 267 |
__stp_relay_create_buf_file_callback(const char *filename, |
| 268 |
struct dentry *parent, |
| 269 |
#ifdef STAPCONF_RELAY_UMODE_T |
| 270 |
umode_t mode, |
| 271 |
#else |
| 272 |
int mode, |
| 273 |
#endif |
| 274 |
struct rchan_buf *buf, |
| 275 |
int *is_global) |
| 276 |
{ |
| 277 |
if (debugfs_p) |
| 278 |
return __stp_debugfs_relay_create_buf_file_callback(filename, parent, mode, buf, is_global); |
| 279 |
if (procfs_p) |
| 280 |
return __stp_procfs_relay_create_buf_file_callback(filename, parent, mode, buf, is_global); |
| 281 |
return NULL; |
| 282 |
} |
| 283 |
|
| 284 |
|
| 285 |
static struct rchan_callbacks __stp_relay_callbacks = { |
| 286 |
.subbuf_start = __stp_relay_subbuf_start_callback, |
| 287 |
.create_buf_file = __stp_relay_create_buf_file_callback, |
| 288 |
.remove_buf_file = __stp_relay_remove_buf_file_callback, |
| 289 |
}; |
| 290 |
|
| 291 |
|
| 292 |
static void _stp_transport_data_fs_start(void) |
| 293 |
{ |
| 294 |
if (atomic_read (&_stp_relay_data.transport_state) == STP_TRANSPORT_INITIALIZED) { |
| 295 |
atomic_set (&_stp_relay_data.transport_state, STP_TRANSPORT_RUNNING); |
| 296 |
/* We're initialized. Now start the timer. */ |
| 297 |
__stp_relay_timer_init(); |
| 298 |
} |
| 299 |
} |
| 300 |
|
| 301 |
static void _stp_transport_data_fs_stop(void) |
| 302 |
{ |
| 303 |
|
| 304 |
if (atomic_read (&_stp_relay_data.transport_state) == STP_TRANSPORT_RUNNING) { |
| 305 |
atomic_set (&_stp_relay_data.transport_state, STP_TRANSPORT_STOPPED); |
| 306 |
STP_TIMER_DELETE_SYNC(&_stp_relay_data.timer); |
| 307 |
dbug_trans(0, "flushing...\n"); |
| 308 |
if (_stp_relay_data.rchan) |
| 309 |
relay_flush(_stp_relay_data.rchan); |
| 310 |
} |
| 311 |
} |
| 312 |
|
| 313 |
static void _stp_transport_data_fs_close(void) |
| 314 |
{ |
| 315 |
_stp_transport_data_fs_stop(); |
| 316 |
if (_stp_relay_data.rchan) { |
| 317 |
relay_close(_stp_relay_data.rchan); |
| 318 |
_stp_relay_data.rchan = NULL; |
| 319 |
} |
| 320 |
} |
| 321 |
|
| 322 |
static int _stp_transport_data_fs_init(void) |
| 323 |
{ |
| 324 |
int rc; |
| 325 |
u64 npages; |
| 326 |
struct sysinfo si; |
| 327 |
|
| 328 |
atomic_set(&_stp_relay_data.transport_state, STP_TRANSPORT_STOPPED); |
| 329 |
_stp_relay_data.overwrite_flag = 0; |
| 330 |
_stp_relay_data.rchan = NULL; |
| 331 |
|
| 332 |
/* Create "trace" file. */ |
| 333 |
npages = _stp_subbuf_size * _stp_nsubbufs; |
| 334 |
npages *= num_online_cpus(); |
| 335 |
npages >>= PAGE_SHIFT; |
| 336 |
si_meminfo(&si); |
| 337 |
#define MB(i) (unsigned long)((i) >> (20 - PAGE_SHIFT)) |
| 338 |
if (npages > (si.freeram + si.bufferram)) { |
| 339 |
errk("Not enough free+buffered memory(%luMB) for log buffer(%luMB)\n", |
| 340 |
MB(si.freeram + si.bufferram), |
| 341 |
MB(npages)); |
| 342 |
rc = -ENOMEM; |
| 343 |
goto err; |
| 344 |
} |
| 345 |
else if (npages > si.freeram) { |
| 346 |
/* exceeds freeram, but below freeram+bufferram */ |
| 347 |
printk(KERN_WARNING |
| 348 |
"log buffer size exceeds free memory(%luMB)\n", |
| 349 |
MB(si.freeram)); |
| 350 |
} |
| 351 |
relay_file_operations_w_owner = relay_file_operations; |
| 352 |
relay_file_operations_w_owner.owner = THIS_MODULE; |
| 353 |
relay_file_operations_w_owner.poll = __stp_relay_file_poll; |
| 354 |
relay_file_operations_w_owner.read = __stp_relay_file_read; |
| 355 |
#if (RELAYFS_CHANNEL_VERSION >= 7) |
| 356 |
_stp_relay_data.rchan = relay_open("trace", _stp_get_module_dir(), |
| 357 |
_stp_subbuf_size, _stp_nsubbufs, |
| 358 |
&__stp_relay_callbacks, NULL); |
| 359 |
#else /* (RELAYFS_CHANNEL_VERSION < 7) */ |
| 360 |
_stp_relay_data.rchan = relay_open("trace", _stp_get_module_dir(), |
| 361 |
_stp_subbuf_size, _stp_nsubbufs, |
| 362 |
&__stp_relay_callbacks); |
| 363 |
#endif /* (RELAYFS_CHANNEL_VERSION < 7) */ |
| 364 |
if (!_stp_relay_data.rchan) { |
| 365 |
rc = -ENOENT; |
| 366 |
errk("%s: relay_open() failed: %d\n", THIS_MODULE->name, rc); |
| 367 |
goto err; |
| 368 |
} |
| 369 |
/* Increment _stp_allocated_memory and _stp_allocated_net_memory to account for buffers |
| 370 |
allocated by relay_open. */ |
| 371 |
{ |
| 372 |
u64 relay_mem; |
| 373 |
|
| 374 |
/* relay_open() invokes relay_open_buf() to allocate this part |
| 375 |
* of memory for online CPUs only. |
| 376 |
*/ |
| 377 |
relay_mem = _stp_subbuf_size * _stp_nsubbufs; |
| 378 |
relay_mem *= num_online_cpus(); |
| 379 |
|
| 380 |
relay_mem += sizeof(struct rchan); |
| 381 |
|
| 382 |
if (sizeof(_stp_relay_data.rchan->buf) == sizeof(struct rchan_buf *)) { |
| 383 |
/* newer kernels (newer than 3.10, for example) */ |
| 384 |
/* this part was allocated by alloc_percpu() inside |
| 385 |
* relay_open(). */ |
| 386 |
relay_mem += sizeof(struct rchan_buf *) * num_possible_cpus(); |
| 387 |
|
| 388 |
} else { |
| 389 |
/* for older kernels like 3.10, the buf pointers are |
| 390 |
* already part of struct rchan as its array member |
| 391 |
* 'struct rchan_buf *buf[NR_CPUS]'. so we do nothing |
| 392 |
* here. |
| 393 |
*/ |
| 394 |
} |
| 395 |
|
| 396 |
_stp_allocated_net_memory += relay_mem; |
| 397 |
_stp_allocated_memory += relay_mem; |
| 398 |
} |
| 399 |
|
| 400 |
dbug_trans(1, "returning 0...\n"); |
| 401 |
atomic_set (&_stp_relay_data.transport_state, STP_TRANSPORT_INITIALIZED); |
| 402 |
|
| 403 |
return 0; |
| 404 |
|
| 405 |
err: |
| 406 |
_stp_transport_data_fs_close(); |
| 407 |
return rc; |
| 408 |
} |
| 409 |
|
| 410 |
|
| 411 |
/** |
| 412 |
* _stp_data_write_reserve - try to reserve size_request bytes |
| 413 |
* @size_request: number of bytes to attempt to reserve |
| 414 |
* @entry: entry is returned here |
| 415 |
* |
| 416 |
* Returns number of bytes reserved, 0 if full. On return, entry |
| 417 |
* will point to allocated opaque pointer. Use |
| 418 |
* _stp_data_entry_data() to get pointer to copy data into. |
| 419 |
* |
| 420 |
* (For this code's purposes, entry is filled in with the actual |
| 421 |
* data pointer, but the caller doesn't know that.) |
| 422 |
*/ |
| 423 |
static size_t |
| 424 |
_stp_data_write_reserve(size_t size_request, void **entry) |
| 425 |
{ |
| 426 |
struct rchan_buf *buf; |
| 427 |
|
| 428 |
if (entry == NULL) |
| 429 |
return -EINVAL; |
| 430 |
|
| 431 |
buf = _stp_get_rchan_subbuf(_stp_relay_data.rchan->buf, |
| 432 |
smp_processor_id()); |
| 433 |
|
| 434 |
/* relay_open() only initializes bufs on the online CPUs at the time |
| 435 |
* of invocation. The online CPUs may change at any time, so the |
| 436 |
* current online CPU might be offline when relay_open() was called. |
| 437 |
* So we must check if the buf for the current CPU is invalid otherwise |
| 438 |
* we may dereference a NULL pointer. |
| 439 |
*/ |
| 440 |
if (unlikely(buf == NULL)) |
| 441 |
return -EINVAL; |
| 442 |
|
| 443 |
if (buf->offset >= buf->chan->subbuf_size) { |
| 444 |
size_request = __stp_relay_switch_subbuf(buf, size_request); |
| 445 |
if (!size_request) |
| 446 |
return 0; |
| 447 |
} else if (buf->offset + size_request > buf->chan->subbuf_size) { |
| 448 |
size_request = buf->chan->subbuf_size - buf->offset; |
| 449 |
} |
| 450 |
*entry = (char*)buf->data + buf->offset; |
| 451 |
buf->offset += size_request; |
| 452 |
|
| 453 |
return size_request; |
| 454 |
} |
| 455 |
|
| 456 |
static unsigned char *_stp_data_entry_data(void *entry) |
| 457 |
{ |
| 458 |
/* Nothing to do here. */ |
| 459 |
return entry; |
| 460 |
} |
| 461 |
|
| 462 |
static int _stp_data_write_commit(void *entry) |
| 463 |
{ |
| 464 |
atomic_set(&_stp_relay_data.wakeup, 1); |
| 465 |
return 0; |
| 466 |
} |