fs/aufs/wkq.c

   1 /*
   2  * Copyright (C) 2005-2017 Junjiro R. Okajima
   3  *
   4  * This program, aufs is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  16  */
  17
  18 /*
  19  * workqueue for asynchronous/super-io operations
  20  * todo: try new dredential scheme
  21  */
  22
  23 #include <linux/module.h>
  24 #include "aufs.h"
  25
  26 /* internal workqueue named AUFS_WKQ_NAME */
  27
  28 static struct workqueue_struct *au_wkq;
  29
  30 struct au_wkinfo {
  31         struct work_struct wk;
  32         struct kobject *kobj;
  33
  34         unsigned int flags; /* see wkq.h */
  35
  36         au_wkq_func_t func;
  37         void *args;
  38
  39 #ifdef CONFIG_LOCKDEP
  40         int dont_check;
  41         struct held_lock **hlock;
  42 #endif
  43
  44         struct completion *comp;
  45 };
  46
  47 /* ---------------------------------------------------------------------- */
  48 /*
  49  * Aufs passes some operations to the workqueue such as the internal copyup.
  50  * This scheme looks rather unnatural for LOCKDEP debugging feature, since the
  51  * job run by workqueue depends upon the locks acquired in the other task.
  52  * Delegating a small operation to the workqueue, aufs passes its lockdep
  53  * information too. And the job in the workqueue restores the info in order to
  54  * pretend as if it acquired those locks. This is just to make LOCKDEP work
  55  * correctly and expectedly.
  56  */
  57
  58 #ifndef CONFIG_LOCKDEP
  59 AuStubInt0(au_wkq_lockdep_alloc, struct au_wkinfo *wkinfo);
  60 AuStubVoid(au_wkq_lockdep_free, struct au_wkinfo *wkinfo);
  61 AuStubVoid(au_wkq_lockdep_pre, struct au_wkinfo *wkinfo);
  62 AuStubVoid(au_wkq_lockdep_post, struct au_wkinfo *wkinfo);
  63 AuStubVoid(au_wkq_lockdep_init, struct au_wkinfo *wkinfo);
  64 #else
  65 static void au_wkq_lockdep_init(struct au_wkinfo *wkinfo)
  66 {
  67         wkinfo->hlock = NULL;
  68         wkinfo->dont_check = 0;
  69 }
  70
  71 /*
  72  * 1: matched
  73  * 0: unmatched
  74  */
  75 static int au_wkq_lockdep_test(struct lock_class_key *key, const char *name)
  76 {
  77         static DEFINE_SPINLOCK(spin);
  78         static struct {
  79                 char *name;
  80                 struct lock_class_key *key;
  81         } a[] = {
  82                 { .name = "&sbinfo->si_rwsem" },
  83                 { .name = "&finfo->fi_rwsem" },
  84                 { .name = "&dinfo->di_rwsem" },
  85                 { .name = "&iinfo->ii_rwsem" }
  86         };
  87         static int set;
  88         int i;
  89
  90         /* lockless read from 'set.' see below */
  91         if (set == ARRAY_SIZE(a)) {
  92                 for (i = 0; i < ARRAY_SIZE(a); i++)
  93                         if (a[i].key == key)
  94                                 goto match;
  95                 goto unmatch;
  96         }
  97
  98         spin_lock(&spin);
  99         if (set)
 100                 for (i = 0; i < ARRAY_SIZE(a); i++)
 101                         if (a[i].key == key) {
 102                                 spin_unlock(&spin);
 103                                 goto match;
 104                         }
 105         for (i = 0; i < ARRAY_SIZE(a); i++) {
 106                 if (a[i].key) {
 107                         if (unlikely(a[i].key == key)) { /* rare but possible */
 108                                 spin_unlock(&spin);
 109                                 goto match;
 110                         } else
 111                                 continue;
 112                 }
 113                 if (strstr(a[i].name, name)) {
 114                         /*
 115                          * the order of these three lines is important for the
 116                          * lockless read above.
 117                          */
 118                         a[i].key = key;
 119                         spin_unlock(&spin);
 120                         set++;
 121                         /* AuDbg("%d, %s\n", set, name); */
 122                         goto match;
 123                 }
 124         }
 125         spin_unlock(&spin);
 126         goto unmatch;
 127
 128 match:
 129         return 1;
 130 unmatch:
 131         return 0;
 132 }
 133
 134 static int au_wkq_lockdep_alloc(struct au_wkinfo *wkinfo)
 135 {
 136         int err, n;
 137         struct task_struct *curr;
 138         struct held_lock **hl, *held_locks, *p;
 139
 140         err = 0;
 141         curr = current;
 142         wkinfo->dont_check = lockdep_recursing(curr);
 143         if (wkinfo->dont_check)
 144                 goto out;
 145         n = curr->lockdep_depth;
 146         if (!n)
 147                 goto out;
 148
 149         err = -ENOMEM;
 150         wkinfo->hlock = kmalloc_array(n + 1, sizeof(*wkinfo->hlock), GFP_NOFS);
 151         if (unlikely(!wkinfo->hlock))
 152                 goto out;
 153
 154         err = 0;
 155 #if 0
 156         if (0 && au_debug_test()) /* left for debugging */
 157                 lockdep_print_held_locks(curr);
 158 #endif
 159         held_locks = curr->held_locks;
 160         hl = wkinfo->hlock;
 161         while (n--) {
 162                 p = held_locks++;
 163                 if (au_wkq_lockdep_test(p->instance->key, p->instance->name))
 164                         *hl++ = p;
 165         }
 166         *hl = NULL;
 167
 168 out:
 169         return err;
 170 }
 171
 172 static void au_wkq_lockdep_free(struct au_wkinfo *wkinfo)
 173 {
 174         kfree(wkinfo->hlock);
 175 }
 176
 177 static void au_wkq_lockdep_pre(struct au_wkinfo *wkinfo)
 178 {
 179         struct held_lock *p, **hl = wkinfo->hlock;
 180         int subclass;
 181
 182         if (wkinfo->dont_check)
 183                 lockdep_off();
 184         if (!hl)
 185                 return;
 186         while ((p = *hl++)) { /* assignment */
 187                 subclass = lockdep_hlock_class(p)->subclass;
 188                 /* AuDbg("%s, %d\n", p->instance->name, subclass); */
 189                 if (p->read)
 190                         rwsem_acquire_read(p->instance, subclass, 0,
 191                                            /*p->acquire_ip*/_RET_IP_);
 192                 else
 193                         rwsem_acquire(p->instance, subclass, 0,
 194                                       /*p->acquire_ip*/_RET_IP_);
 195         }
 196 }
 197
 198 static void au_wkq_lockdep_post(struct au_wkinfo *wkinfo)
 199 {
 200         struct held_lock *p, **hl = wkinfo->hlock;
 201
 202         if (wkinfo->dont_check)
 203                 lockdep_on();
 204         if (!hl)
 205                 return;
 206         while ((p = *hl++)) /* assignment */
 207                 rwsem_release(p->instance, 0, /*p->acquire_ip*/_RET_IP_);
 208 }
 209 #endif
 210
 211 static void wkq_func(struct work_struct *wk)
 212 {
 213         struct au_wkinfo *wkinfo = container_of(wk, struct au_wkinfo, wk);
 214
 215         AuDebugOn(!uid_eq(current_fsuid(), GLOBAL_ROOT_UID));
 216         AuDebugOn(rlimit(RLIMIT_FSIZE) != RLIM_INFINITY);
 217
 218         au_wkq_lockdep_pre(wkinfo);
 219         wkinfo->func(wkinfo->args);
 220         au_wkq_lockdep_post(wkinfo);
 221         if (au_ftest_wkq(wkinfo->flags, WAIT))
 222                 complete(wkinfo->comp);
 223         else {
 224                 kobject_put(wkinfo->kobj);
 225                 module_put(THIS_MODULE); /* todo: ?? */
 226                 kfree(wkinfo);
 227         }
 228 }
 229
 230 /*
 231  * Since struct completion is large, try allocating it dynamically.
 232  */
 233 #if 1 /* defined(CONFIG_4KSTACKS) || defined(AuTest4KSTACKS) */
 234 #define AuWkqCompDeclare(name)  struct completion *comp = NULL
 235
 236 static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
 237 {
 238         *comp = kmalloc(sizeof(**comp), GFP_NOFS);
 239         if (*comp) {
 240                 init_completion(*comp);
 241                 wkinfo->comp = *comp;
 242                 return 0;
 243         }
 244         return -ENOMEM;
 245 }
 246
 247 static void au_wkq_comp_free(struct completion *comp)
 248 {
 249         kfree(comp);
 250 }
 251
 252 #else
 253
 254 /* no braces */
 255 #define AuWkqCompDeclare(name) \
 256         DECLARE_COMPLETION_ONSTACK(_ ## name); \
 257         struct completion *comp = &_ ## name
 258
 259 static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
 260 {
 261         wkinfo->comp = *comp;
 262         return 0;
 263 }
 264
 265 static void au_wkq_comp_free(struct completion *comp __maybe_unused)
 266 {
 267         /* empty */
 268 }
 269 #endif /* 4KSTACKS */
 270
 271 static void au_wkq_run(struct au_wkinfo *wkinfo)
 272 {
 273         if (au_ftest_wkq(wkinfo->flags, NEST)) {
 274                 if (au_wkq_test()) {
 275                         AuWarn1("wkq from wkq, unless silly-rename on NFS,"
 276                                 " due to a dead dir by UDBA?\n");
 277                         AuDebugOn(au_ftest_wkq(wkinfo->flags, WAIT));
 278                 }
 279         } else
 280                 au_dbg_verify_kthread();
 281
 282         if (au_ftest_wkq(wkinfo->flags, WAIT)) {
 283                 INIT_WORK_ONSTACK(&wkinfo->wk, wkq_func);
 284                 queue_work(au_wkq, &wkinfo->wk);
 285         } else {
 286                 INIT_WORK(&wkinfo->wk, wkq_func);
 287                 schedule_work(&wkinfo->wk);
 288         }
 289 }
 290
 291 /*
 292  * Be careful. It is easy to make deadlock happen.
 293  * processA: lock, wkq and wait
 294  * processB: wkq and wait, lock in wkq
 295  * --> deadlock
 296  */
 297 int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args)
 298 {
 299         int err;
 300         AuWkqCompDeclare(comp);
 301         struct au_wkinfo wkinfo = {
 302                 .flags  = flags,
 303                 .func   = func,
 304                 .args   = args
 305         };
 306
 307         err = au_wkq_comp_alloc(&wkinfo, &comp);
 308         if (unlikely(err))
 309                 goto out;
 310         err = au_wkq_lockdep_alloc(&wkinfo);
 311         if (unlikely(err))
 312                 goto out_comp;
 313         if (!err) {
 314                 au_wkq_run(&wkinfo);
 315                 /* no timeout, no interrupt */
 316                 wait_for_completion(wkinfo.comp);
 317         }
 318         au_wkq_lockdep_free(&wkinfo);
 319
 320 out_comp:
 321         au_wkq_comp_free(comp);
 322 out:
 323         destroy_work_on_stack(&wkinfo.wk);
 324         return err;
 325 }
 326
 327 /*
 328  * Note: dget/dput() in func for aufs dentries are not supported. It will be a
 329  * problem in a concurrent umounting.
 330  */
 331 int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
 332                   unsigned int flags)
 333 {
 334         int err;
 335         struct au_wkinfo *wkinfo;
 336
 337         atomic_inc(&au_sbi(sb)->si_nowait.nw_len);
 338
 339         /*
 340          * wkq_func() must free this wkinfo.
 341          * it highly depends upon the implementation of workqueue.
 342          */
 343         err = 0;
 344         wkinfo = kmalloc(sizeof(*wkinfo), GFP_NOFS);
 345         if (wkinfo) {
 346                 wkinfo->kobj = &au_sbi(sb)->si_kobj;
 347                 wkinfo->flags = flags & ~AuWkq_WAIT;
 348                 wkinfo->func = func;
 349                 wkinfo->args = args;
 350                 wkinfo->comp = NULL;
 351                 au_wkq_lockdep_init(wkinfo);
 352                 kobject_get(wkinfo->kobj);
 353                 __module_get(THIS_MODULE); /* todo: ?? */
 354
 355                 au_wkq_run(wkinfo);
 356         } else {
 357                 err = -ENOMEM;
 358                 au_nwt_done(&au_sbi(sb)->si_nowait);
 359         }
 360
 361         return err;
 362 }
 363
 364 /* ---------------------------------------------------------------------- */
 365
 366 void au_nwt_init(struct au_nowait_tasks *nwt)
 367 {
 368         atomic_set(&nwt->nw_len, 0);
 369         /* smp_mb(); */ /* atomic_set */
 370         init_waitqueue_head(&nwt->nw_wq);
 371 }
 372
 373 void au_wkq_fin(void)
 374 {
 375         destroy_workqueue(au_wkq);
 376 }
 377
 378 int __init au_wkq_init(void)
 379 {
 380         int err;
 381
 382         err = 0;
 383         au_wkq = alloc_workqueue(AUFS_WKQ_NAME, 0, WQ_DFL_ACTIVE);
 384         if (IS_ERR(au_wkq))
 385                 err = PTR_ERR(au_wkq);
 386         else if (!au_wkq)
 387                 err = -ENOMEM;
 388
 389         return err;
 390 }