server/change.c

   1 /*
   2  * Server-side change notification management
   3  *
   4  * Copyright (C) 1998 Alexandre Julliard
   5  * Copyright (C) 2006 Mike McCormack
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  20  */
  21
  22 #include "config.h"
  23 #include "wine/port.h"
  24
  25 #include <assert.h>
  26 #include <fcntl.h>
  27 #include <stdio.h>
  28 #include <stdlib.h>
  29 #include <signal.h>
  30 #include <sys/stat.h>
  31 #include <sys/types.h>
  32 #include <limits.h>
  33 #include <dirent.h>
  34 #include <errno.h>
  35 #ifdef HAVE_SYS_ERRNO_H
  36 #include <sys/errno.h>
  37 #endif
  38
  39 #include "ntstatus.h"
  40 #define WIN32_NO_STATUS
  41 #include "windef.h"
  42
  43 #include "file.h"
  44 #include "handle.h"
  45 #include "thread.h"
  46 #include "request.h"
  47 #include "process.h"
  48 #include "security.h"
  49 #include "winternl.h"
  50
  51 /* dnotify support */
  52
  53 #ifdef linux
  54 #ifndef F_NOTIFY
  55 #define F_NOTIFY 1026
  56 #define DN_ACCESS       0x00000001      /* File accessed */
  57 #define DN_MODIFY       0x00000002      /* File modified */
  58 #define DN_CREATE       0x00000004      /* File created */
  59 #define DN_DELETE       0x00000008      /* File removed */
  60 #define DN_RENAME       0x00000010      /* File renamed */
  61 #define DN_ATTRIB       0x00000020      /* File changed attributes */
  62 #define DN_MULTISHOT    0x80000000      /* Don't remove notifier */
  63 #endif
  64 #endif
  65
  66 /* inotify support */
  67
  68 #ifdef HAVE_SYS_INOTIFY_H
  69 #include <sys/inotify.h>
  70 #define USE_INOTIFY
  71 #elif defined(__linux__) && defined(__i386__)
  72
  73 #define SYS_inotify_init        291
  74 #define SYS_inotify_add_watch   292
  75 #define SYS_inotify_rm_watch    293
  76
  77 struct inotify_event {
  78     int           wd;
  79     unsigned int  mask;
  80     unsigned int  cookie;
  81     unsigned int  len;
  82     char          name[1];
  83 };
  84
  85 #define IN_ACCESS        0x00000001
  86 #define IN_MODIFY        0x00000002
  87 #define IN_ATTRIB        0x00000004
  88 #define IN_CLOSE_WRITE   0x00000008
  89 #define IN_CLOSE_NOWRITE 0x00000010
  90 #define IN_OPEN          0x00000020
  91 #define IN_MOVED_FROM    0x00000040
  92 #define IN_MOVED_TO      0x00000080
  93 #define IN_CREATE        0x00000100
  94 #define IN_DELETE        0x00000200
  95 #define IN_DELETE_SELF   0x00000400
  96
  97 #define IN_ISDIR         0x40000000
  98
  99 static inline int inotify_init( void )
 100 {
 101     return syscall( SYS_inotify_init );
 102 }
 103
 104 static inline int inotify_add_watch( int fd, const char *name, unsigned int mask )
 105 {
 106     return syscall( SYS_inotify_add_watch, fd, name, mask );
 107 }
 108
 109 static inline int inotify_rm_watch( int fd, int wd )
 110 {
 111     return syscall( SYS_inotify_rm_watch, fd, wd );
 112 }
 113
 114 #define USE_INOTIFY
 115
 116 #endif
 117
 118 struct inode;
 119
 120 static void free_inode( struct inode *inode );
 121
 122 static struct fd *inotify_fd;
 123
 124 struct change_record {
 125     struct list entry;
 126     unsigned int cookie;
 127     struct filesystem_event event;
 128 };
 129
 130 struct dir
 131 {
 132     struct object  obj;      /* object header */
 133     struct fd     *fd;       /* file descriptor to the directory */
 134     mode_t         mode;     /* file stat.st_mode */
 135     uid_t          uid;      /* file stat.st_uid */
 136     struct list    entry;    /* entry in global change notifications list */
 137     unsigned int   filter;   /* notification filter */
 138     int            notified; /* SIGIO counter */
 139     int            want_data; /* return change data */
 140     int            subtree;  /* do we want to watch subdirectories? */
 141     struct list    change_records;   /* data for the change */
 142     struct list    in_entry; /* entry in the inode dirs list */
 143     struct inode  *inode;    /* inode of the associated directory */
 144 };
 145
 146 static struct fd *dir_get_fd( struct object *obj );
 147 static struct security_descriptor *dir_get_sd( struct object *obj );
 148 static int dir_set_sd( struct object *obj, const struct security_descriptor *sd,
 149                        unsigned int set_info );
 150 static void dir_dump( struct object *obj, int verbose );
 151 static void dir_destroy( struct object *obj );
 152
 153 static const struct object_ops dir_ops =
 154 {
 155     sizeof(struct dir),       /* size */
 156     dir_dump,                 /* dump */
 157     no_get_type,              /* get_type */
 158     add_queue,                /* add_queue */
 159     remove_queue,             /* remove_queue */
 160     default_fd_signaled,      /* signaled */
 161     no_satisfied,             /* satisfied */
 162     no_signal,                /* signal */
 163     dir_get_fd,               /* get_fd */
 164     default_fd_map_access,    /* map_access */
 165     dir_get_sd,               /* get_sd */
 166     dir_set_sd,               /* set_sd */
 167     no_lookup_name,           /* lookup_name */
 168     no_open_file,             /* open_file */
 169     fd_close_handle,          /* close_handle */
 170     dir_destroy               /* destroy */
 171 };
 172
 173 static int dir_get_poll_events( struct fd *fd );
 174 static enum server_fd_type dir_get_fd_type( struct fd *fd );
 175
 176 static const struct fd_ops dir_fd_ops =
 177 {
 178     dir_get_poll_events,         /* get_poll_events */
 179     default_poll_event,          /* poll_event */
 180     no_flush,                    /* flush */
 181     dir_get_fd_type,             /* get_fd_type */
 182     default_fd_ioctl,            /* ioctl */
 183     default_fd_queue_async,      /* queue_async */
 184     default_fd_reselect_async,   /* reselect_async */
 185     default_fd_cancel_async      /* cancel_async */
 186 };
 187
 188 static struct list change_list = LIST_INIT(change_list);
 189
 190 static void dnotify_adjust_changes( struct dir *dir )
 191 {
 192 #if defined(F_SETSIG) && defined(F_NOTIFY)
 193     int fd = get_unix_fd( dir->fd );
 194     unsigned int filter = dir->filter;
 195     unsigned int val;
 196     if ( 0 > fcntl( fd, F_SETSIG, SIGIO) )
 197         return;
 198
 199     val = DN_MULTISHOT;
 200     if (filter & FILE_NOTIFY_CHANGE_FILE_NAME)
 201         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 202     if (filter & FILE_NOTIFY_CHANGE_DIR_NAME)
 203         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 204     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 205         val |= DN_ATTRIB;
 206     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 207         val |= DN_MODIFY;
 208     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 209         val |= DN_MODIFY;
 210     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 211         val |= DN_ACCESS;
 212     if (filter & FILE_NOTIFY_CHANGE_CREATION)
 213         val |= DN_CREATE;
 214     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 215         val |= DN_ATTRIB;
 216     fcntl( fd, F_NOTIFY, val );
 217 #endif
 218 }
 219
 220 /* insert change in the global list */
 221 static inline void insert_change( struct dir *dir )
 222 {
 223     sigset_t sigset;
 224
 225     sigemptyset( &sigset );
 226     sigaddset( &sigset, SIGIO );
 227     sigprocmask( SIG_BLOCK, &sigset, NULL );
 228     list_add_head( &change_list, &dir->entry );
 229     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 230 }
 231
 232 /* remove change from the global list */
 233 static inline void remove_change( struct dir *dir )
 234 {
 235     sigset_t sigset;
 236
 237     sigemptyset( &sigset );
 238     sigaddset( &sigset, SIGIO );
 239     sigprocmask( SIG_BLOCK, &sigset, NULL );
 240     list_remove( &dir->entry );
 241     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 242 }
 243
 244 static void dir_dump( struct object *obj, int verbose )
 245 {
 246     struct dir *dir = (struct dir *)obj;
 247     assert( obj->ops == &dir_ops );
 248     fprintf( stderr, "Dirfile fd=%p filter=%08x\n", dir->fd, dir->filter );
 249 }
 250
 251 /* enter here directly from SIGIO signal handler */
 252 void do_change_notify( int unix_fd )
 253 {
 254     struct dir *dir;
 255
 256     /* FIXME: this is O(n) ... probably can be improved */
 257     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 258     {
 259         if (get_unix_fd( dir->fd ) != unix_fd) continue;
 260         interlocked_xchg_add( &dir->notified, 1 );
 261         break;
 262     }
 263 }
 264
 265 /* SIGIO callback, called synchronously with the poll loop */
 266 void sigio_callback(void)
 267 {
 268     struct dir *dir;
 269
 270     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 271     {
 272         if (interlocked_xchg( &dir->notified, 0 ))
 273             fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_NOTIFY_ENUM_DIR );
 274     }
 275 }
 276
 277 static struct fd *dir_get_fd( struct object *obj )
 278 {
 279     struct dir *dir = (struct dir *)obj;
 280     assert( obj->ops == &dir_ops );
 281     return (struct fd *)grab_object( dir->fd );
 282 }
 283
 284 static int get_dir_unix_fd( struct dir *dir )
 285 {
 286     return get_unix_fd( dir->fd );
 287 }
 288
 289 static struct security_descriptor *dir_get_sd( struct object *obj )
 290 {
 291     struct dir *dir = (struct dir *)obj;
 292     int unix_fd;
 293     struct stat st;
 294     struct security_descriptor *sd;
 295     assert( obj->ops == &dir_ops );
 296
 297     unix_fd = get_dir_unix_fd( dir );
 298
 299     if (unix_fd == -1 || fstat( unix_fd, &st ) == -1)
 300         return obj->sd;
 301
 302     /* mode and uid the same? if so, no need to re-generate security descriptor */
 303     if (obj->sd &&
 304         (st.st_mode & (S_IRWXU|S_IRWXO)) == (dir->mode & (S_IRWXU|S_IRWXO)) &&
 305         (st.st_uid == dir->uid))
 306         return obj->sd;
 307
 308     sd = mode_to_sd( st.st_mode,
 309                      security_unix_uid_to_sid( st.st_uid ),
 310                      token_get_primary_group( current->process->token ));
 311     if (!sd) return obj->sd;
 312
 313     dir->mode = st.st_mode;
 314     dir->uid = st.st_uid;
 315     free( obj->sd );
 316     obj->sd = sd;
 317     return sd;
 318 }
 319
 320 static int dir_set_sd( struct object *obj, const struct security_descriptor *sd,
 321                        unsigned int set_info )
 322 {
 323     struct dir *dir = (struct dir *)obj;
 324     const SID *owner;
 325     struct stat st;
 326     mode_t mode;
 327     int unix_fd;
 328
 329     assert( obj->ops == &dir_ops );
 330
 331     unix_fd = get_dir_unix_fd( dir );
 332
 333     if (unix_fd == -1 || fstat( unix_fd, &st ) == -1) return 1;
 334
 335     if (set_info & OWNER_SECURITY_INFORMATION)
 336     {
 337         owner = sd_get_owner( sd );
 338         if (!owner)
 339         {
 340             set_error( STATUS_INVALID_SECURITY_DESCR );
 341             return 0;
 342         }
 343         if (!obj->sd || !security_equal_sid( owner, sd_get_owner( obj->sd ) ))
 344         {
 345             /* FIXME: get Unix uid and call fchown */
 346         }
 347     }
 348     else if (obj->sd)
 349         owner = sd_get_owner( obj->sd );
 350     else
 351         owner = token_get_user( current->process->token );
 352
 353     if (set_info & DACL_SECURITY_INFORMATION)
 354     {
 355         /* keep the bits that we don't map to access rights in the ACL */
 356         mode = st.st_mode & (S_ISUID|S_ISGID|S_ISVTX);
 357         mode |= sd_to_mode( sd, owner );
 358
 359         if (((st.st_mode ^ mode) & (S_IRWXU|S_IRWXG|S_IRWXO)) && fchmod( unix_fd, mode ) == -1)
 360         {
 361             file_set_error();
 362             return 0;
 363         }
 364     }
 365     return 1;
 366 }
 367
 368 static struct change_record *get_first_change_record( struct dir *dir )
 369 {
 370     struct list *ptr = list_head( &dir->change_records );
 371     if (!ptr) return NULL;
 372     list_remove( ptr );
 373     return LIST_ENTRY( ptr, struct change_record, entry );
 374 }
 375
 376 static void dir_destroy( struct object *obj )
 377 {
 378     struct change_record *record;
 379     struct dir *dir = (struct dir *)obj;
 380     assert (obj->ops == &dir_ops);
 381
 382     if (dir->filter)
 383         remove_change( dir );
 384
 385     if (dir->inode)
 386     {
 387         list_remove( &dir->in_entry );
 388         free_inode( dir->inode );
 389     }
 390
 391     while ((record = get_first_change_record( dir ))) free( record );
 392
 393     release_object( dir->fd );
 394
 395     if (inotify_fd && list_empty( &change_list ))
 396     {
 397         release_object( inotify_fd );
 398         inotify_fd = NULL;
 399     }
 400 }
 401
 402 struct dir *get_dir_obj( struct process *process, obj_handle_t handle, unsigned int access )
 403 {
 404     return (struct dir *)get_handle_obj( process, handle, access, &dir_ops );
 405 }
 406
 407 static int dir_get_poll_events( struct fd *fd )
 408 {
 409     return 0;
 410 }
 411
 412 static enum server_fd_type dir_get_fd_type( struct fd *fd )
 413 {
 414     return FD_TYPE_DIR;
 415 }
 416
 417 #ifdef USE_INOTIFY
 418
 419 #define HASH_SIZE 31
 420
 421 struct inode {
 422     struct list ch_entry;    /* entry in the children list */
 423     struct list children;    /* children of this inode */
 424     struct inode *parent;    /* parent of this inode */
 425     struct list dirs;        /* directory handles watching this inode */
 426     struct list ino_entry;   /* entry in the inode hash */
 427     struct list wd_entry;    /* entry in the watch descriptor hash */
 428     dev_t dev;               /* device number */
 429     ino_t ino;               /* device's inode number */
 430     int wd;                  /* inotify's watch descriptor */
 431     char *name;              /* basename name of the inode */
 432 };
 433
 434 static struct list inode_hash[ HASH_SIZE ];
 435 static struct list wd_hash[ HASH_SIZE ];
 436
 437 static int inotify_add_dir( char *path, unsigned int filter );
 438
 439 static struct inode *inode_from_wd( int wd )
 440 {
 441     struct list *bucket = &wd_hash[ wd % HASH_SIZE ];
 442     struct inode *inode;
 443
 444     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, wd_entry )
 445         if (inode->wd == wd)
 446             return inode;
 447
 448     return NULL;
 449 }
 450
 451 static inline struct list *get_hash_list( dev_t dev, ino_t ino )
 452 {
 453     return &inode_hash[ (ino ^ dev) % HASH_SIZE ];
 454 }
 455
 456 static struct inode *find_inode( dev_t dev, ino_t ino )
 457 {
 458     struct list *bucket = get_hash_list( dev, ino );
 459     struct inode *inode;
 460
 461     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, ino_entry )
 462         if (inode->ino == ino && inode->dev == dev)
 463              return inode;
 464
 465     return NULL;
 466 }
 467
 468 static struct inode *create_inode( dev_t dev, ino_t ino )
 469 {
 470     struct inode *inode;
 471
 472     inode = malloc( sizeof *inode );
 473     if (inode)
 474     {
 475         list_init( &inode->children );
 476         list_init( &inode->dirs );
 477         inode->ino = ino;
 478         inode->dev = dev;
 479         inode->wd = -1;
 480         inode->parent = NULL;
 481         inode->name = NULL;
 482         list_add_tail( get_hash_list( dev, ino ), &inode->ino_entry );
 483     }
 484     return inode;
 485 }
 486
 487 static struct inode *get_inode( dev_t dev, ino_t ino )
 488 {
 489     struct inode *inode;
 490
 491     inode = find_inode( dev, ino );
 492     if (inode)
 493         return inode;
 494     return create_inode( dev, ino );
 495 }
 496
 497 static void inode_set_wd( struct inode *inode, int wd )
 498 {
 499     if (inode->wd != -1)
 500         list_remove( &inode->wd_entry );
 501     inode->wd = wd;
 502     list_add_tail( &wd_hash[ wd % HASH_SIZE ], &inode->wd_entry );
 503 }
 504
 505 static void inode_set_name( struct inode *inode, const char *name )
 506 {
 507     free (inode->name);
 508     inode->name = name ? strdup( name ) : NULL;
 509 }
 510
 511 static void free_inode( struct inode *inode )
 512 {
 513     int subtree = 0, watches = 0;
 514     struct inode *tmp, *next;
 515     struct dir *dir;
 516
 517     LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 518     {
 519         subtree |= dir->subtree;
 520         watches++;
 521     }
 522
 523     if (!subtree && !inode->parent)
 524     {
 525         LIST_FOR_EACH_ENTRY_SAFE( tmp, next, &inode->children,
 526                                   struct inode, ch_entry )
 527         {
 528             assert( tmp != inode );
 529             assert( tmp->parent == inode );
 530             free_inode( tmp );
 531         }
 532     }
 533
 534     if (watches)
 535         return;
 536
 537     if (inode->parent)
 538         list_remove( &inode->ch_entry );
 539
 540     /* disconnect remaining children from the parent */
 541     LIST_FOR_EACH_ENTRY_SAFE( tmp, next, &inode->children, struct inode, ch_entry )
 542     {
 543         list_remove( &tmp->ch_entry );
 544         tmp->parent = NULL;
 545     }
 546
 547     if (inode->wd != -1)
 548     {
 549         inotify_rm_watch( get_unix_fd( inotify_fd ), inode->wd );
 550         list_remove( &inode->wd_entry );
 551     }
 552     list_remove( &inode->ino_entry );
 553
 554     free( inode->name );
 555     free( inode );
 556 }
 557
 558 static struct inode *inode_add( struct inode *parent,
 559                                 dev_t dev, ino_t ino, const char *name )
 560 {
 561     struct inode *inode;
 562
 563     inode = get_inode( dev, ino );
 564     if (!inode)
 565         return NULL;
 566
 567     if (!inode->parent)
 568     {
 569         list_add_tail( &parent->children, &inode->ch_entry );
 570         inode->parent = parent;
 571         assert( inode != parent );
 572     }
 573     inode_set_name( inode, name );
 574
 575     return inode;
 576 }
 577
 578 static struct inode *inode_from_name( struct inode *inode, const char *name )
 579 {
 580     struct inode *i;
 581
 582     LIST_FOR_EACH_ENTRY( i, &inode->children, struct inode, ch_entry )
 583         if (i->name && !strcmp( i->name, name ))
 584             return i;
 585     return NULL;
 586 }
 587
 588 static int inotify_get_poll_events( struct fd *fd );
 589 static void inotify_poll_event( struct fd *fd, int event );
 590
 591 static const struct fd_ops inotify_fd_ops =
 592 {
 593     inotify_get_poll_events,     /* get_poll_events */
 594     inotify_poll_event,          /* poll_event */
 595     NULL,                        /* flush */
 596     NULL,                        /* get_fd_type */
 597     NULL,                        /* ioctl */
 598     NULL,                        /* queue_async */
 599     NULL,                        /* reselect_async */
 600     NULL,                        /* cancel_async */
 601 };
 602
 603 static int inotify_get_poll_events( struct fd *fd )
 604 {
 605     return POLLIN;
 606 }
 607
 608 static void inotify_do_change_notify( struct dir *dir, unsigned int action,
 609                                       unsigned int cookie, const char *relpath )
 610 {
 611     struct change_record *record;
 612
 613     assert( dir->obj.ops == &dir_ops );
 614
 615     if (dir->want_data)
 616     {
 617         size_t len = strlen(relpath);
 618         record = malloc( offsetof(struct change_record, event.name[len]) );
 619         if (!record)
 620             return;
 621
 622         record->cookie = cookie;
 623         record->event.action = action;
 624         memcpy( record->event.name, relpath, len );
 625         record->event.len = len;
 626
 627         list_add_tail( &dir->change_records, &record->entry );
 628     }
 629
 630     fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_ALERTED );
 631 }
 632
 633 static unsigned int filter_from_event( struct inotify_event *ie )
 634 {
 635     unsigned int filter = 0;
 636
 637     if (ie->mask & (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE))
 638         filter |= FILE_NOTIFY_CHANGE_FILE_NAME | FILE_NOTIFY_CHANGE_DIR_NAME;
 639     if (ie->mask & IN_MODIFY)
 640         filter |= FILE_NOTIFY_CHANGE_SIZE | FILE_NOTIFY_CHANGE_LAST_WRITE;
 641     if (ie->mask & IN_ATTRIB)
 642         filter |= FILE_NOTIFY_CHANGE_ATTRIBUTES | FILE_NOTIFY_CHANGE_SECURITY;
 643     if (ie->mask & IN_ACCESS)
 644         filter |= FILE_NOTIFY_CHANGE_LAST_ACCESS;
 645     if (ie->mask & IN_CREATE)
 646         filter |= FILE_NOTIFY_CHANGE_CREATION;
 647
 648     if (ie->mask & IN_ISDIR)
 649         filter &= ~FILE_NOTIFY_CHANGE_FILE_NAME;
 650     else
 651         filter &= ~FILE_NOTIFY_CHANGE_DIR_NAME;
 652
 653     return filter;
 654 }
 655
 656 /* scan up the parent directories for watches */
 657 static unsigned int filter_from_inode( struct inode *inode, int is_parent )
 658 {
 659     unsigned int filter = 0;
 660     struct dir *dir;
 661
 662     /* combine filters from parents watching subtrees */
 663     while (inode)
 664     {
 665         LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 666             if (dir->subtree || !is_parent)
 667                 filter |= dir->filter;
 668         is_parent = 1;
 669         inode = inode->parent;
 670     }
 671
 672     return filter;
 673 }
 674
 675 static char *inode_get_path( struct inode *inode, int sz )
 676 {
 677     struct list *head;
 678     char *path;
 679     int len;
 680
 681     if (!inode)
 682         return NULL;
 683
 684     head = list_head( &inode->dirs );
 685     if (head)
 686     {
 687         int unix_fd = get_unix_fd( LIST_ENTRY( head, struct dir, in_entry )->fd );
 688         path = malloc ( 32 + sz );
 689         if (path)
 690             sprintf( path, "/proc/self/fd/%u/", unix_fd );
 691         return path;
 692     }
 693
 694     if (!inode->name)
 695         return NULL;
 696
 697     len = strlen( inode->name );
 698     path = inode_get_path( inode->parent, sz + len + 1 );
 699     if (!path)
 700         return NULL;
 701
 702     strcat( path, inode->name );
 703     strcat( path, "/" );
 704
 705     return path;
 706 }
 707
 708 static void inode_check_dir( struct inode *parent, const char *name )
 709 {
 710     char *path;
 711     unsigned int filter;
 712     struct inode *inode;
 713     struct stat st;
 714     int wd = -1;
 715
 716     path = inode_get_path( parent, strlen(name) );
 717     if (!path)
 718         return;
 719
 720     strcat( path, name );
 721
 722     if (stat( path, &st ) < 0)
 723         goto end;
 724
 725     filter = filter_from_inode( parent, 1 );
 726     if (!filter)
 727         goto end;
 728
 729     inode = inode_add( parent, st.st_dev, st.st_ino, name );
 730     if (!inode || inode->wd != -1)
 731         goto end;
 732
 733     wd = inotify_add_dir( path, filter );
 734     if (wd != -1)
 735         inode_set_wd( inode, wd );
 736     else
 737         free_inode( inode );
 738
 739 end:
 740     free( path );
 741 }
 742
 743 static int prepend( char **path, const char *segment )
 744 {
 745     int extra;
 746     char *p;
 747
 748     extra = strlen( segment ) + 1;
 749     if (*path)
 750     {
 751         int len = strlen( *path ) + 1;
 752         p = realloc( *path, len + extra );
 753         if (!p) return 0;
 754         memmove( &p[ extra ], p, len );
 755         p[ extra - 1 ] = '/';
 756         memcpy( p, segment, extra - 1 );
 757     }
 758     else
 759     {
 760         p = malloc( extra );
 761         if (!p) return 0;
 762         memcpy( p, segment, extra );
 763     }
 764
 765     *path = p;
 766
 767     return 1;
 768 }
 769
 770 static void inotify_notify_all( struct inotify_event *ie )
 771 {
 772     unsigned int filter, action;
 773     struct inode *inode, *i;
 774     char *path = NULL;
 775     struct dir *dir;
 776
 777     inode = inode_from_wd( ie->wd );
 778     if (!inode)
 779     {
 780         fprintf( stderr, "no inode matches %d\n", ie->wd);
 781         return;
 782     }
 783
 784     filter = filter_from_event( ie );
 785
 786     if (ie->mask & IN_CREATE)
 787     {
 788         if (ie->mask & IN_ISDIR)
 789             inode_check_dir( inode, ie->name );
 790
 791         action = FILE_ACTION_ADDED;
 792     }
 793     else if (ie->mask & IN_DELETE)
 794         action = FILE_ACTION_REMOVED;
 795     else if (ie->mask & IN_MOVED_FROM)
 796         action = FILE_ACTION_RENAMED_OLD_NAME;
 797     else if (ie->mask & IN_MOVED_TO)
 798         action = FILE_ACTION_RENAMED_NEW_NAME;
 799     else
 800         action = FILE_ACTION_MODIFIED;
 801
 802     /*
 803      * Work our way up the inode hierarchy
 804      *  extending the relative path as we go
 805      *  and notifying all recursive watches.
 806      */
 807     if (!prepend( &path, ie->name ))
 808         return;
 809
 810     for (i = inode; i; i = i->parent)
 811     {
 812         LIST_FOR_EACH_ENTRY( dir, &i->dirs, struct dir, in_entry )
 813             if ((filter & dir->filter) && (i==inode || dir->subtree))
 814                 inotify_do_change_notify( dir, action, ie->cookie, path );
 815
 816         if (!i->name || !prepend( &path, i->name ))
 817             break;
 818     }
 819
 820     free( path );
 821
 822     if (ie->mask & IN_DELETE)
 823     {
 824         i = inode_from_name( inode, ie->name );
 825         if (i)
 826             free_inode( i );
 827     }
 828 }
 829
 830 static void inotify_poll_event( struct fd *fd, int event )
 831 {
 832     int r, ofs, unix_fd;
 833     char buffer[0x1000];
 834     struct inotify_event *ie;
 835
 836     unix_fd = get_unix_fd( fd );
 837     r = read( unix_fd, buffer, sizeof buffer );
 838     if (r < 0)
 839     {
 840         fprintf(stderr,"inotify_poll_event(): inotify read failed!\n");
 841         return;
 842     }
 843
 844     for( ofs = 0; ofs < r - offsetof(struct inotify_event, name); )
 845     {
 846         ie = (struct inotify_event*) &buffer[ofs];
 847         if (!ie->len)
 848             break;
 849         ofs += offsetof( struct inotify_event, name[ie->len] );
 850         if (ofs > r) break;
 851         inotify_notify_all( ie );
 852     }
 853 }
 854
 855 static inline struct fd *create_inotify_fd( void )
 856 {
 857     int unix_fd;
 858
 859     unix_fd = inotify_init();
 860     if (unix_fd<0)
 861         return NULL;
 862     return create_anonymous_fd( &inotify_fd_ops, unix_fd, NULL, 0 );
 863 }
 864
 865 static int map_flags( unsigned int filter )
 866 {
 867     unsigned int mask;
 868
 869     /* always watch these so we can track subdirectories in recursive watches */
 870     mask = (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE | IN_DELETE_SELF);
 871
 872     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 873         mask |= IN_ATTRIB;
 874     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 875         mask |= IN_MODIFY;
 876     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 877         mask |= IN_MODIFY;
 878     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 879         mask |= IN_ACCESS;
 880     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 881         mask |= IN_ATTRIB;
 882
 883     return mask;
 884 }
 885
 886 static int inotify_add_dir( char *path, unsigned int filter )
 887 {
 888     int wd = inotify_add_watch( get_unix_fd( inotify_fd ),
 889                                 path, map_flags( filter ) );
 890     if (wd != -1)
 891         set_fd_events( inotify_fd, POLLIN );
 892     return wd;
 893 }
 894
 895 static int init_inotify( void )
 896 {
 897     int i;
 898
 899     if (inotify_fd)
 900         return 1;
 901
 902     inotify_fd = create_inotify_fd();
 903     if (!inotify_fd)
 904         return 0;
 905
 906     for (i=0; i<HASH_SIZE; i++)
 907     {
 908         list_init( &inode_hash[i] );
 909         list_init( &wd_hash[i] );
 910     }
 911
 912     return 1;
 913 }
 914
 915 static int inotify_adjust_changes( struct dir *dir )
 916 {
 917     unsigned int filter;
 918     struct inode *inode;
 919     struct stat st;
 920     char path[32];
 921     int wd, unix_fd;
 922
 923     if (!inotify_fd)
 924         return 0;
 925
 926     unix_fd = get_unix_fd( dir->fd );
 927
 928     inode = dir->inode;
 929     if (!inode)
 930     {
 931         /* check if this fd is already being watched */
 932         if (-1 == fstat( unix_fd, &st ))
 933             return 0;
 934
 935         inode = get_inode( st.st_dev, st.st_ino );
 936         if (!inode)
 937             inode = create_inode( st.st_dev, st.st_ino );
 938         if (!inode)
 939             return 0;
 940         list_add_tail( &inode->dirs, &dir->in_entry );
 941         dir->inode = inode;
 942     }
 943
 944     filter = filter_from_inode( inode, 0 );
 945
 946     sprintf( path, "/proc/self/fd/%u", unix_fd );
 947     wd = inotify_add_dir( path, filter );
 948     if (wd == -1) return 0;
 949
 950     inode_set_wd( inode, wd );
 951
 952     return 1;
 953 }
 954
 955 static char *get_basename( const char *link )
 956 {
 957     char *buffer, *name = NULL;
 958     int r, n = 0x100;
 959
 960     while (1)
 961     {
 962         buffer = malloc( n );
 963         if (!buffer) return NULL;
 964
 965         r = readlink( link, buffer, n );
 966         if (r < 0)
 967             break;
 968
 969         if (r < n)
 970         {
 971             name = buffer;
 972             break;
 973         }
 974         free( buffer );
 975         n *= 2;
 976     }
 977
 978     if (name)
 979     {
 980         while (r > 0 && name[ r - 1 ] == '/' )
 981             r--;
 982         name[ r ] = 0;
 983
 984         name = strrchr( name, '/' );
 985         if (name)
 986             name = strdup( &name[1] );
 987     }
 988
 989     free( buffer );
 990     return name;
 991 }
 992
 993 static int dir_add_to_existing_notify( struct dir *dir )
 994 {
 995     struct inode *inode, *parent;
 996     unsigned int filter = 0;
 997     struct stat st, st_new;
 998     char link[35], *name;
 999     int wd, unix_fd;
1000
1001     if (!inotify_fd)
1002         return 0;
1003
1004     unix_fd = get_unix_fd( dir->fd );
1005
1006     /* check if it's in the list of inodes we want to watch */
1007     if (-1 == fstat( unix_fd, &st_new ))
1008         return 0;
1009     inode = find_inode( st_new.st_dev, st_new.st_ino );
1010     if (inode)
1011         return 0;
1012
1013     /* lookup the parent */
1014     sprintf( link, "/proc/self/fd/%u/..", unix_fd );
1015     if (-1 == stat( link, &st ))
1016         return 0;
1017
1018     /*
1019      * If there's no parent, stop.  We could keep going adding
1020      *  ../ to the path until we hit the root of the tree or
1021      *  find a recursively watched ancestor.
1022      * Assume it's too expensive to search up the tree for now.
1023      */
1024     parent = find_inode( st.st_dev, st.st_ino );
1025     if (!parent)
1026         return 0;
1027
1028     if (parent->wd == -1)
1029         return 0;
1030
1031     filter = filter_from_inode( parent, 1 );
1032     if (!filter)
1033         return 0;
1034
1035     sprintf( link, "/proc/self/fd/%u", unix_fd );
1036     name = get_basename( link );
1037     if (!name)
1038         return 0;
1039     inode = inode_add( parent, st_new.st_dev, st_new.st_ino, name );
1040     free( name );
1041     if (!inode)
1042         return 0;
1043
1044     /* Couldn't find this inode at the start of the function, must be new */
1045     assert( inode->wd == -1 );
1046
1047     wd = inotify_add_dir( link, filter );
1048     if (wd != -1)
1049         inode_set_wd( inode, wd );
1050
1051     return 1;
1052 }
1053
1054 #else
1055
1056 static int init_inotify( void )
1057 {
1058     return 0;
1059 }
1060
1061 static int inotify_adjust_changes( struct dir *dir )
1062 {
1063     return 0;
1064 }
1065
1066 static void free_inode( struct inode *inode )
1067 {
1068     assert( 0 );
1069 }
1070
1071 static int dir_add_to_existing_notify( struct dir *dir )
1072 {
1073     return 0;
1074 }
1075
1076 #endif  /* USE_INOTIFY */
1077
1078 struct object *create_dir_obj( struct fd *fd, unsigned int access, mode_t mode )
1079 {
1080     struct dir *dir;
1081
1082     dir = alloc_object( &dir_ops );
1083     if (!dir)
1084         return NULL;
1085
1086     list_init( &dir->change_records );
1087     dir->filter = 0;
1088     dir->notified = 0;
1089     dir->want_data = 0;
1090     dir->inode = NULL;
1091     grab_object( fd );
1092     dir->fd = fd;
1093     dir->mode = mode;
1094     dir->uid  = ~(uid_t)0;
1095     set_fd_user( fd, &dir_fd_ops, &dir->obj );
1096
1097     dir_add_to_existing_notify( dir );
1098
1099     return &dir->obj;
1100 }
1101
1102 /* enable change notifications for a directory */
1103 DECL_HANDLER(read_directory_changes)
1104 {
1105     struct dir *dir;
1106     struct async *async;
1107
1108     if (!req->filter)
1109     {
1110         set_error(STATUS_INVALID_PARAMETER);
1111         return;
1112     }
1113
1114     dir = get_dir_obj( current->process, req->async.handle, 0 );
1115     if (!dir)
1116         return;
1117
1118     /* requests don't timeout */
1119     if (!(async = fd_queue_async( dir->fd, &req->async, ASYNC_TYPE_WAIT ))) goto end;
1120
1121     /* assign it once */
1122     if (!dir->filter)
1123     {
1124         init_inotify();
1125         insert_change( dir );
1126         dir->filter = req->filter;
1127         dir->subtree = req->subtree;
1128         dir->want_data = req->want_data;
1129     }
1130
1131     /* if there's already a change in the queue, send it */
1132     if (!list_empty( &dir->change_records ))
1133         fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_ALERTED );
1134
1135     /* setup the real notification */
1136     if (!inotify_adjust_changes( dir ))
1137         dnotify_adjust_changes( dir );
1138
1139     release_object( async );
1140     set_error(STATUS_PENDING);
1141
1142 end:
1143     release_object( dir );
1144 }
1145
1146 DECL_HANDLER(read_change)
1147 {
1148     struct change_record *record, *next;
1149     struct dir *dir;
1150     struct list events;
1151     char *data, *event;
1152     int size = 0;
1153
1154     dir = get_dir_obj( current->process, req->handle, 0 );
1155     if (!dir)
1156         return;
1157
1158     list_init( &events );
1159     list_move_tail( &events, &dir->change_records );
1160     release_object( dir );
1161
1162     if (list_empty( &events ))
1163     {
1164         set_error( STATUS_NO_DATA_DETECTED );
1165         return;
1166     }
1167
1168     LIST_FOR_EACH_ENTRY( record, &events, struct change_record, entry )
1169     {
1170         size += (offsetof(struct filesystem_event, name[record->event.len])
1171                 + sizeof(int)-1) / sizeof(int) * sizeof(int);
1172     }
1173
1174     if (size > get_reply_max_size())
1175         set_error( STATUS_BUFFER_TOO_SMALL );
1176     else if ((data = mem_alloc( size )) != NULL)
1177     {
1178         event = data;
1179         LIST_FOR_EACH_ENTRY( record, &events, struct change_record, entry )
1180         {
1181             data_size_t len = offsetof( struct filesystem_event, name[record->event.len] );
1182
1183             /* FIXME: rename events are sometimes reported as delete/create */
1184             if (record->event.action == FILE_ACTION_RENAMED_OLD_NAME)
1185             {
1186                 struct list *elem = list_next( &events, &record->entry );
1187                 if (elem)
1188                     next = LIST_ENTRY(elem, struct change_record, entry);
1189
1190                 if (elem && next->cookie == record->cookie)
1191                     next->cookie = 0;
1192                 else
1193                     record->event.action = FILE_ACTION_REMOVED;
1194             }
1195             else if (record->event.action == FILE_ACTION_RENAMED_NEW_NAME && record->cookie)
1196                 record->event.action = FILE_ACTION_ADDED;
1197
1198             memcpy( event, &record->event, len );
1199             event += len;
1200             if (len % sizeof(int))
1201             {
1202                 memset( event, 0, sizeof(int) - len % sizeof(int) );
1203                 event += sizeof(int) - len % sizeof(int);
1204             }
1205         }
1206         set_reply_data_ptr( data, size );
1207     }
1208
1209     LIST_FOR_EACH_ENTRY_SAFE( record, next, &events, struct change_record, entry )
1210     {
1211         list_remove( &record->entry );
1212         free( record );
1213     }
1214 }