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., 59 Temple Place, Suite 330, Boston, MA  02111-1307  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 "winternl.h"
  48
  49 /* dnotify support */
  50
  51 #ifdef linux
  52 #ifndef F_NOTIFY
  53 #define F_NOTIFY 1026
  54 #define DN_ACCESS       0x00000001      /* File accessed */
  55 #define DN_MODIFY       0x00000002      /* File modified */
  56 #define DN_CREATE       0x00000004      /* File created */
  57 #define DN_DELETE       0x00000008      /* File removed */
  58 #define DN_RENAME       0x00000010      /* File renamed */
  59 #define DN_ATTRIB       0x00000020      /* File changed attibutes */
  60 #define DN_MULTISHOT    0x80000000      /* Don't remove notifier */
  61 #endif
  62 #endif
  63
  64 /* inotify support */
  65
  66 #if defined(__linux__) && defined(__i386__)
  67
  68 #define SYS_inotify_init        291
  69 #define SYS_inotify_add_watch   292
  70 #define SYS_inotify_rm_watch    293
  71
  72 struct inotify_event {
  73     int           wd;
  74     unsigned int  mask;
  75     unsigned int  cookie;
  76     unsigned int  len;
  77     char          name[1];
  78 };
  79
  80 #define IN_ACCESS        0x00000001
  81 #define IN_MODIFY        0x00000002
  82 #define IN_ATTRIB        0x00000004
  83 #define IN_CLOSE_WRITE   0x00000008
  84 #define IN_CLOSE_NOWRITE 0x00000010
  85 #define IN_OPEN          0x00000020
  86 #define IN_MOVED_FROM    0x00000040
  87 #define IN_MOVED_TO      0x00000080
  88 #define IN_CREATE        0x00000100
  89 #define IN_DELETE        0x00000200
  90 #define IN_DELETE_SELF   0x00000400
  91
  92 static inline int inotify_init( void )
  93 {
  94     int ret;
  95     __asm__ __volatile__( "int $0x80"
  96                           : "=a" (ret)
  97                           : "0" (SYS_inotify_init));
  98     if (ret<0) { errno = -ret; ret = -1; }
  99     return ret;
 100 }
 101
 102 static inline int inotify_add_watch( int fd, const char *name, unsigned int mask )
 103 {
 104     int ret;
 105     __asm__ __volatile__( "pushl %%ebx;\n\t"
 106                           "movl %2,%%ebx;\n\t"
 107                           "int $0x80;\n\t"
 108                           "popl %%ebx"
 109                           : "=a" (ret) : "0" (SYS_inotify_add_watch),
 110                             "r" (fd), "c" (name), "d" (mask) );
 111     if (ret<0) { errno = -ret; ret = -1; }
 112     return ret;
 113 }
 114
 115 static inline int inotify_remove_watch( int fd, int wd )
 116 {
 117     int ret;
 118     __asm__ __volatile__( "pushl %%ebx;\n\t"
 119                           "movl %2,%%ebx;\n\t"
 120                           "int $0x80;\n\t"
 121                           "popl %%ebx"
 122                           : "=a" (ret) : "0" (SYS_inotify_rm_watch),
 123                             "r" (fd), "c" (wd) );
 124     if (ret<0) { errno = -ret; ret = -1; }
 125     return ret;
 126 }
 127
 128 #define USE_INOTIFY
 129
 130 #endif
 131
 132 struct inode;
 133
 134 static void free_inode( struct inode *inode );
 135
 136 static struct fd *inotify_fd;
 137
 138 struct change_record {
 139     struct list entry;
 140     int action;
 141     int len;
 142     char name[1];
 143 };
 144
 145 struct dir
 146 {
 147     struct object  obj;      /* object header */
 148     struct fd     *fd;       /* file descriptor to the directory */
 149     struct list    entry;    /* entry in global change notifications list */
 150     struct event  *event;
 151     unsigned int   filter;   /* notification filter */
 152     int            notified; /* SIGIO counter */
 153     int            want_data; /* return change data */
 154     long           signaled; /* the file changed */
 155     int            subtree;  /* do we want to watch subdirectories? */
 156     struct list    change_q; /* change readers */
 157     struct list    change_records;   /* data for the change */
 158     struct list    in_entry; /* entry in the inode dirs list */
 159     struct inode  *inode;    /* inode of the associated directory */
 160 };
 161
 162 static struct fd *dir_get_fd( struct object *obj );
 163 static unsigned int dir_map_access( struct object *obj, unsigned int access );
 164 static void dir_dump( struct object *obj, int verbose );
 165 static void dir_destroy( struct object *obj );
 166 static int dir_signaled( struct object *obj, struct thread *thread );
 167
 168 static const struct object_ops dir_ops =
 169 {
 170     sizeof(struct dir),       /* size */
 171     dir_dump,                 /* dump */
 172     add_queue,                /* add_queue */
 173     remove_queue,             /* remove_queue */
 174     dir_signaled,             /* signaled */
 175     no_satisfied,             /* satisfied */
 176     no_signal,                /* signal */
 177     dir_get_fd,               /* get_fd */
 178     dir_map_access,           /* map_access */
 179     no_lookup_name,           /* lookup_name */
 180     no_close_handle,          /* close_handle */
 181     dir_destroy               /* destroy */
 182 };
 183
 184 static int dir_get_poll_events( struct fd *fd );
 185 static int dir_get_info( struct fd *fd );
 186 static void dir_cancel_async( struct fd *fd );
 187
 188 static const struct fd_ops dir_fd_ops =
 189 {
 190     dir_get_poll_events,      /* get_poll_events */
 191     default_poll_event,       /* poll_event */
 192     no_flush,                 /* flush */
 193     dir_get_info,             /* get_file_info */
 194     default_fd_queue_async,   /* queue_async */
 195     dir_cancel_async          /* cancel_async */
 196 };
 197
 198 static struct list change_list = LIST_INIT(change_list);
 199
 200 static void dnotify_adjust_changes( struct dir *dir )
 201 {
 202 #if defined(F_SETSIG) && defined(F_NOTIFY)
 203     int fd = get_unix_fd( dir->fd );
 204     unsigned int filter = dir->filter;
 205     unsigned int val;
 206     if ( 0 > fcntl( fd, F_SETSIG, SIGIO) )
 207         return;
 208
 209     val = DN_MULTISHOT;
 210     if (filter & FILE_NOTIFY_CHANGE_FILE_NAME)
 211         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 212     if (filter & FILE_NOTIFY_CHANGE_DIR_NAME)
 213         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 214     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 215         val |= DN_ATTRIB;
 216     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 217         val |= DN_MODIFY;
 218     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 219         val |= DN_MODIFY;
 220     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 221         val |= DN_ACCESS;
 222     if (filter & FILE_NOTIFY_CHANGE_CREATION)
 223         val |= DN_CREATE;
 224     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 225         val |= DN_ATTRIB;
 226     fcntl( fd, F_NOTIFY, val );
 227 #endif
 228 }
 229
 230 /* insert change in the global list */
 231 static inline void insert_change( struct dir *dir )
 232 {
 233     sigset_t sigset;
 234
 235     sigemptyset( &sigset );
 236     sigaddset( &sigset, SIGIO );
 237     sigprocmask( SIG_BLOCK, &sigset, NULL );
 238     list_add_head( &change_list, &dir->entry );
 239     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 240 }
 241
 242 /* remove change from the global list */
 243 static inline void remove_change( struct dir *dir )
 244 {
 245     sigset_t sigset;
 246
 247     sigemptyset( &sigset );
 248     sigaddset( &sigset, SIGIO );
 249     sigprocmask( SIG_BLOCK, &sigset, NULL );
 250     list_remove( &dir->entry );
 251     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 252 }
 253
 254 static void dir_dump( struct object *obj, int verbose )
 255 {
 256     struct dir *dir = (struct dir *)obj;
 257     assert( obj->ops == &dir_ops );
 258     fprintf( stderr, "Dirfile fd=%p event=%p filter=%08x\n",
 259              dir->fd, dir->event, dir->filter );
 260 }
 261
 262 static int dir_signaled( struct object *obj, struct thread *thread )
 263 {
 264     struct dir *dir = (struct dir *)obj;
 265     assert (obj->ops == &dir_ops);
 266     return (dir->event == NULL) && dir->signaled;
 267 }
 268
 269 /* enter here directly from SIGIO signal handler */
 270 void do_change_notify( int unix_fd )
 271 {
 272     struct dir *dir;
 273
 274     /* FIXME: this is O(n) ... probably can be improved */
 275     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 276     {
 277         if (get_unix_fd( dir->fd ) != unix_fd) continue;
 278         interlocked_xchg_add( &dir->notified, 1 );
 279         break;
 280     }
 281 }
 282
 283 static void dir_signal_changed( struct dir *dir )
 284 {
 285     if (dir->event)
 286         set_event( dir->event );
 287     else
 288         wake_up( &dir->obj, 0 );
 289 }
 290
 291 /* SIGIO callback, called synchronously with the poll loop */
 292 void sigio_callback(void)
 293 {
 294     struct dir *dir;
 295
 296     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 297     {
 298         long count = interlocked_xchg( &dir->notified, 0 );
 299         if (count)
 300         {
 301             dir->signaled += count;
 302             if (dir->signaled == count)  /* was it 0? */
 303                 dir_signal_changed( dir );
 304         }
 305     }
 306 }
 307
 308 static struct fd *dir_get_fd( struct object *obj )
 309 {
 310     struct dir *dir = (struct dir *)obj;
 311     assert( obj->ops == &dir_ops );
 312     return (struct fd *)grab_object( dir->fd );
 313 }
 314
 315 static unsigned int dir_map_access( struct object *obj, unsigned int access )
 316 {
 317     if (access & GENERIC_READ)    access |= FILE_GENERIC_READ;
 318     if (access & GENERIC_WRITE)   access |= FILE_GENERIC_WRITE;
 319     if (access & GENERIC_EXECUTE) access |= FILE_GENERIC_EXECUTE;
 320     if (access & GENERIC_ALL)     access |= FILE_ALL_ACCESS;
 321     return access & ~(GENERIC_READ | GENERIC_WRITE | GENERIC_EXECUTE | GENERIC_ALL);
 322 }
 323
 324 static struct change_record *get_first_change_record( struct dir *dir )
 325 {
 326     struct list *ptr = list_head( &dir->change_records );
 327     if (!ptr) return NULL;
 328     list_remove( ptr );
 329     return LIST_ENTRY( ptr, struct change_record, entry );
 330 }
 331
 332 static void dir_destroy( struct object *obj )
 333 {
 334     struct change_record *record;
 335     struct dir *dir = (struct dir *)obj;
 336     assert (obj->ops == &dir_ops);
 337
 338     if (dir->filter)
 339         remove_change( dir );
 340
 341     if (dir->inode)
 342     {
 343         list_remove( &dir->in_entry );
 344         free_inode( dir->inode );
 345     }
 346
 347     async_terminate_queue( &dir->change_q, STATUS_CANCELLED );
 348     while ((record = get_first_change_record( dir ))) free( record );
 349
 350     if (dir->event)
 351     {
 352         set_event( dir->event );
 353         release_object( dir->event );
 354     }
 355     release_object( dir->fd );
 356
 357     if (inotify_fd && list_empty( &change_list ))
 358     {
 359         release_object( inotify_fd );
 360         inotify_fd = NULL;
 361     }
 362 }
 363
 364 static struct dir *
 365 get_dir_obj( struct process *process, obj_handle_t handle, unsigned int access )
 366 {
 367     return (struct dir *)get_handle_obj( process, handle, access, &dir_ops );
 368 }
 369
 370 static int dir_get_poll_events( struct fd *fd )
 371 {
 372     return 0;
 373 }
 374
 375 static int dir_get_info( struct fd *fd )
 376 {
 377     return 0;
 378 }
 379
 380 static void dir_cancel_async( struct fd *fd )
 381 {
 382     struct dir *dir = (struct dir *) get_fd_user( fd );
 383     async_terminate_queue( &dir->change_q, STATUS_CANCELLED );
 384 }
 385
 386
 387 #ifdef USE_INOTIFY
 388
 389 #define HASH_SIZE 31
 390
 391 struct inode {
 392     struct list ch_entry;    /* entry in the children list */
 393     struct list children;    /* children of this inode */
 394     struct inode *parent;    /* parent of this inode */
 395     struct list dirs;        /* directory handles watching this inode */
 396     struct list ino_entry;   /* entry in the inode hash */
 397     struct list wd_entry;    /* entry in the watch descriptor hash */
 398     dev_t dev;               /* device number */
 399     ino_t ino;               /* device's inode number */
 400     int wd;                  /* inotify's watch descriptor */
 401     char *name;              /* basename name of the inode */
 402 };
 403
 404 struct list inode_hash[ HASH_SIZE ];
 405 struct list wd_hash[ HASH_SIZE ];
 406
 407 static int inotify_add_dir( char *path, unsigned int filter );
 408
 409 static struct inode *inode_from_wd( int wd )
 410 {
 411     struct list *bucket = &wd_hash[ wd % HASH_SIZE ];
 412     struct inode *inode;
 413
 414     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, wd_entry )
 415         if (inode->wd == wd)
 416             return inode;
 417
 418     return NULL;
 419 }
 420
 421 static inline struct list *get_hash_list( dev_t dev, ino_t ino )
 422 {
 423     return &inode_hash[ (ino ^ dev) % HASH_SIZE ];
 424 }
 425
 426 static struct inode *find_inode( dev_t dev, ino_t ino )
 427 {
 428     struct list *bucket = get_hash_list( dev, ino );
 429     struct inode *inode;
 430
 431     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, ino_entry )
 432         if (inode->ino == ino && inode->dev == dev)
 433              return inode;
 434
 435     return NULL;
 436 }
 437
 438 static struct inode *create_inode( dev_t dev, ino_t ino )
 439 {
 440     struct inode *inode;
 441
 442     inode = malloc( sizeof *inode );
 443     if (inode)
 444     {
 445         list_init( &inode->children );
 446         list_init( &inode->dirs );
 447         inode->ino = ino;
 448         inode->dev = dev;
 449         inode->wd = -1;
 450         inode->parent = NULL;
 451         inode->name = NULL;
 452         list_add_tail( get_hash_list( dev, ino ), &inode->ino_entry );
 453     }
 454     return inode;
 455 }
 456
 457 static struct inode *get_inode( dev_t dev, ino_t ino )
 458 {
 459     struct inode *inode;
 460
 461     inode = find_inode( dev, ino );
 462     if (inode)
 463         return inode;
 464     return create_inode( dev, ino );
 465 }
 466
 467 static void inode_set_wd( struct inode *inode, int wd )
 468 {
 469     if (inode->wd != -1)
 470         list_remove( &inode->wd_entry );
 471     inode->wd = wd;
 472     list_add_tail( &wd_hash[ wd % HASH_SIZE ], &inode->wd_entry );
 473 }
 474
 475 static void inode_set_name( struct inode *inode, const char *name )
 476 {
 477     if (inode->name)
 478         free (inode->name);
 479     inode->name = name ? strdup( name ) : NULL;
 480 }
 481
 482 static void free_inode( struct inode *inode )
 483 {
 484     int subtree = 0, watches = 0;
 485     struct dir *dir;
 486
 487     LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 488     {
 489         subtree |= dir->subtree;
 490         watches++;
 491     }
 492
 493     if (!subtree && !inode->parent)
 494     {
 495         struct inode *tmp, *next;
 496         LIST_FOR_EACH_ENTRY_SAFE( tmp, next, &inode->children,
 497                                   struct inode, ch_entry )
 498         {
 499             assert( tmp != inode );
 500             assert( tmp->parent == inode );
 501             free_inode( tmp );
 502         }
 503     }
 504
 505     if (watches)
 506         return;
 507
 508     if (inode->parent)
 509         list_remove( &inode->ch_entry );
 510
 511     if (inode->wd != -1)
 512     {
 513         inotify_remove_watch( get_unix_fd( inotify_fd ), inode->wd );
 514         list_remove( &inode->wd_entry );
 515     }
 516     list_remove( &inode->ino_entry );
 517
 518     free( inode->name );
 519     free( inode );
 520 }
 521
 522 static struct inode *inode_add( struct inode *parent,
 523                                 dev_t dev, ino_t ino, const char *name )
 524 {
 525     struct inode *inode;
 526
 527     inode = get_inode( dev, ino );
 528     if (!inode)
 529         return NULL;
 530
 531     if (!inode->parent)
 532     {
 533         list_add_tail( &parent->children, &inode->ch_entry );
 534         inode->parent = parent;
 535         assert( inode != parent );
 536     }
 537     inode_set_name( inode, name );
 538
 539     return inode;
 540 }
 541
 542 static struct inode *inode_from_name( struct inode *inode, const char *name )
 543 {
 544     struct inode *i;
 545
 546     LIST_FOR_EACH_ENTRY( i, &inode->children, struct inode, ch_entry )
 547         if (i->name && !strcmp( i->name, name ))
 548             return i;
 549     return NULL;
 550 }
 551
 552 static int inotify_get_poll_events( struct fd *fd );
 553 static void inotify_poll_event( struct fd *fd, int event );
 554 static int inotify_get_info( struct fd *fd );
 555
 556 static const struct fd_ops inotify_fd_ops =
 557 {
 558     inotify_get_poll_events,  /* get_poll_events */
 559     inotify_poll_event,       /* poll_event */
 560     no_flush,                 /* flush */
 561     inotify_get_info,         /* get_file_info */
 562     default_fd_queue_async,   /* queue_async */
 563     default_fd_cancel_async,  /* cancel_async */
 564 };
 565
 566 static int inotify_get_poll_events( struct fd *fd )
 567 {
 568     return POLLIN;
 569 }
 570
 571 static void inotify_do_change_notify( struct dir *dir, unsigned int action,
 572                                       const char *relpath )
 573 {
 574     struct change_record *record;
 575
 576     assert( dir->obj.ops == &dir_ops );
 577
 578     if (dir->want_data)
 579     {
 580         size_t len = strlen(relpath);
 581         record = malloc( offsetof(struct change_record, name[len]) );
 582         if (!record)
 583             return;
 584
 585         record->action = action;
 586         memcpy( record->name, relpath, len );
 587         record->len = len;
 588
 589         list_add_tail( &dir->change_records, &record->entry );
 590     }
 591
 592     if (!list_empty( &dir->change_q ))
 593         async_terminate_head( &dir->change_q, STATUS_ALERTED );
 594     else
 595     {
 596         dir->signaled++;
 597         dir_signal_changed( dir );
 598     }
 599 }
 600
 601 static unsigned int filter_from_event( struct inotify_event *ie )
 602 {
 603     unsigned int filter = 0;
 604
 605     if (ie->mask & (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE))
 606         filter |= FILE_NOTIFY_CHANGE_FILE_NAME | FILE_NOTIFY_CHANGE_DIR_NAME;
 607     if (ie->mask & IN_MODIFY)
 608         filter |= FILE_NOTIFY_CHANGE_SIZE | FILE_NOTIFY_CHANGE_LAST_WRITE;
 609     if (ie->mask & IN_ATTRIB)
 610         filter |= FILE_NOTIFY_CHANGE_ATTRIBUTES | FILE_NOTIFY_CHANGE_SECURITY;
 611     if (ie->mask & IN_ACCESS)
 612         filter |= FILE_NOTIFY_CHANGE_LAST_ACCESS;
 613     if (ie->mask & IN_CREATE)
 614         filter |= FILE_NOTIFY_CHANGE_CREATION;
 615
 616     return filter;
 617 }
 618
 619 /* scan up the parent directories for watches */
 620 static unsigned int filter_from_inode( struct inode *inode, int is_parent )
 621 {
 622     unsigned int filter = 0;
 623     struct dir *dir;
 624
 625     /* combine filters from parents watching subtrees */
 626     while (inode)
 627     {
 628         LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 629             if (dir->subtree || !is_parent)
 630                 filter |= dir->filter;
 631         is_parent = 1;
 632         inode = inode->parent;
 633     }
 634
 635     return filter;
 636 }
 637
 638 static char *inode_get_path( struct inode *inode, int sz )
 639 {
 640     struct list *head;
 641     char *path;
 642     int len;
 643
 644     if (!inode)
 645         return NULL;
 646
 647     head = list_head( &inode->dirs );
 648     if (head)
 649     {
 650         int unix_fd = get_unix_fd( LIST_ENTRY( head, struct dir, in_entry )->fd );
 651         path = malloc ( 32 + sz );
 652         if (path)
 653             sprintf( path, "/proc/self/fd/%u/", unix_fd );
 654         return path;
 655     }
 656
 657     if (!inode->name)
 658         return NULL;
 659
 660     len = strlen( inode->name );
 661     path = inode_get_path( inode->parent, sz + len + 1 );
 662     if (!path)
 663         return NULL;
 664
 665     strcat( path, inode->name );
 666     strcat( path, "/" );
 667
 668     return path;
 669 }
 670
 671 static int inode_check_dir( struct inode *parent, const char *name )
 672 {
 673     char *path;
 674     unsigned int filter;
 675     struct inode *inode;
 676     struct stat st;
 677     int wd = -1, r = -1;
 678
 679     path = inode_get_path( parent, strlen(name) );
 680     if (!path)
 681         return r;
 682
 683     strcat( path, name );
 684
 685     r = stat( path, &st );
 686     if (r < 0) goto end;
 687
 688     if (!S_ISDIR(st.st_mode))
 689     {
 690         r = 0;
 691         goto end;
 692     }
 693
 694     r = 1;
 695
 696     filter = filter_from_inode( parent, 1 );
 697     if (!filter)
 698         goto end;
 699
 700     inode = inode_add( parent, st.st_dev, st.st_ino, name );
 701     if (!inode || inode->wd != -1)
 702         goto end;
 703
 704     wd = inotify_add_dir( path, filter );
 705     if (wd != -1)
 706         inode_set_wd( inode, wd );
 707     else
 708         free_inode( inode );
 709
 710 end:
 711     free( path );
 712     return r;
 713 }
 714
 715 static int prepend( char **path, const char *segment )
 716 {
 717     int extra;
 718     char *p;
 719
 720     extra = strlen( segment ) + 1;
 721     if (*path)
 722     {
 723         int len = strlen( *path ) + 1;
 724         p = realloc( *path, len + extra );
 725         if (!p) return 0;
 726         memmove( &p[ extra ], p, len );
 727         p[ extra - 1 ] = '/';
 728         memcpy( p, segment, extra - 1 );
 729     }
 730     else
 731     {
 732         p = malloc( extra );
 733         if (!p) return 0;
 734         memcpy( p, segment, extra );
 735     }
 736
 737     *path = p;
 738
 739     return 1;
 740 }
 741
 742 static void inotify_notify_all( struct inotify_event *ie )
 743 {
 744     unsigned int filter, action;
 745     struct inode *inode, *i;
 746     char *path = NULL;
 747     struct dir *dir;
 748
 749     inode = inode_from_wd( ie->wd );
 750     if (!inode)
 751     {
 752         fprintf( stderr, "no inode matches %d\n", ie->wd);
 753         return;
 754     }
 755
 756     filter = filter_from_event( ie );
 757
 758     if (ie->mask & IN_CREATE)
 759     {
 760         switch (inode_check_dir( inode, ie->name ))
 761         {
 762         case 1:
 763             filter &= ~FILE_NOTIFY_CHANGE_FILE_NAME;
 764             break;
 765         case 0:
 766             filter &= ~FILE_NOTIFY_CHANGE_DIR_NAME;
 767             break;
 768         default:
 769             break;
 770             /* Maybe the file disappeared before we could check it? */
 771         }
 772         action = FILE_ACTION_ADDED;
 773     }
 774     else if (ie->mask & IN_DELETE)
 775         action = FILE_ACTION_REMOVED;
 776     else
 777         action = FILE_ACTION_MODIFIED;
 778
 779     /*
 780      * Work our way up the inode hierarchy
 781      *  extending the relative path as we go
 782      *  and notifying all recursive watches.
 783      */
 784     if (!prepend( &path, ie->name ))
 785         return;
 786
 787     for (i = inode; i; i = i->parent)
 788     {
 789         LIST_FOR_EACH_ENTRY( dir, &i->dirs, struct dir, in_entry )
 790             if ((filter & dir->filter) && (i==inode || dir->subtree))
 791                 inotify_do_change_notify( dir, action, path );
 792
 793         if (!i->name || !prepend( &path, i->name ))
 794             break;
 795     }
 796
 797     free( path );
 798
 799     if (ie->mask & IN_DELETE)
 800     {
 801         i = inode_from_name( inode, ie->name );
 802         if (i)
 803             free_inode( i );
 804     }
 805 }
 806
 807 static void inotify_poll_event( struct fd *fd, int event )
 808 {
 809     int r, ofs, unix_fd;
 810     char buffer[0x1000];
 811     struct inotify_event *ie;
 812
 813     unix_fd = get_unix_fd( fd );
 814     r = read( unix_fd, buffer, sizeof buffer );
 815     if (r < 0)
 816     {
 817         fprintf(stderr,"inotify_poll_event(): inotify read failed!\n");
 818         return;
 819     }
 820
 821     for( ofs = 0; ofs < r - offsetof(struct inotify_event, name); )
 822     {
 823         ie = (struct inotify_event*) &buffer[ofs];
 824         if (!ie->len)
 825             break;
 826         ofs += offsetof( struct inotify_event, name[ie->len] );
 827         if (ofs > r) break;
 828         inotify_notify_all( ie );
 829     }
 830 }
 831
 832 static int inotify_get_info( struct fd *fd )
 833 {
 834     return 0;
 835 }
 836
 837 static inline struct fd *create_inotify_fd( void )
 838 {
 839     int unix_fd;
 840
 841     unix_fd = inotify_init();
 842     if (unix_fd<0)
 843         return NULL;
 844     return create_anonymous_fd( &inotify_fd_ops, unix_fd, NULL );
 845 }
 846
 847 static int map_flags( unsigned int filter )
 848 {
 849     unsigned int mask;
 850
 851     /* always watch these so we can track subdirectories in recursive watches */
 852     mask = (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE | IN_DELETE_SELF);
 853
 854     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 855         mask |= IN_ATTRIB;
 856     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 857         mask |= IN_MODIFY;
 858     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 859         mask |= IN_MODIFY;
 860     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 861         mask |= IN_ACCESS;
 862     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 863         mask |= IN_ATTRIB;
 864
 865     return mask;
 866 }
 867
 868 static int inotify_add_dir( char *path, unsigned int filter )
 869 {
 870     int wd = inotify_add_watch( get_unix_fd( inotify_fd ),
 871                                 path, map_flags( filter ) );
 872     if (wd != -1)
 873         set_fd_events( inotify_fd, POLLIN );
 874     return wd;
 875 }
 876
 877 static int init_inotify( void )
 878 {
 879     int i;
 880
 881     if (inotify_fd)
 882         return 1;
 883
 884     inotify_fd = create_inotify_fd();
 885     if (!inotify_fd)
 886         return 0;
 887
 888     for (i=0; i<HASH_SIZE; i++)
 889     {
 890         list_init( &inode_hash[i] );
 891         list_init( &wd_hash[i] );
 892     }
 893
 894     return 1;
 895 }
 896
 897 static int inotify_adjust_changes( struct dir *dir )
 898 {
 899     unsigned int filter;
 900     struct inode *inode;
 901     struct stat st;
 902     char path[32];
 903     int wd, unix_fd;
 904
 905     if (!inotify_fd)
 906         return 0;
 907
 908     unix_fd = get_unix_fd( dir->fd );
 909
 910     inode = dir->inode;
 911     if (!inode)
 912     {
 913         /* check if this fd is already being watched */
 914         if (-1 == fstat( unix_fd, &st ))
 915             return 0;
 916
 917         inode = get_inode( st.st_dev, st.st_ino );
 918         if (!inode)
 919             inode = create_inode( st.st_dev, st.st_ino );
 920         if (!inode)
 921             return 0;
 922         list_add_tail( &inode->dirs, &dir->in_entry );
 923         dir->inode = inode;
 924     }
 925
 926     filter = filter_from_inode( inode, 0 );
 927
 928     sprintf( path, "/proc/self/fd/%u", unix_fd );
 929     wd = inotify_add_dir( path, filter );
 930     if (wd == -1) return 0;
 931
 932     inode_set_wd( inode, wd );
 933
 934     return 1;
 935 }
 936
 937 static char *get_basename( const char *link )
 938 {
 939     char *buffer, *name = NULL;
 940     int r, n = 0x100;
 941
 942     while (1)
 943     {
 944         buffer = malloc( n );
 945         if (!buffer) return NULL;
 946
 947         r = readlink( link, buffer, n );
 948         if (r < 0)
 949             break;
 950
 951         if (r < n)
 952         {
 953             name = buffer;
 954             break;
 955         }
 956         free( buffer );
 957         n *= 2;
 958     }
 959
 960     if (name)
 961     {
 962         while (r > 0 && name[ r - 1 ] == '/' )
 963             r--;
 964         name[ r ] = 0;
 965
 966         name = strrchr( name, '/' );
 967         if (name)
 968             name = strdup( &name[1] );
 969     }
 970
 971     free( buffer );
 972     return name;
 973 }
 974
 975 static int dir_add_to_existing_notify( struct dir *dir )
 976 {
 977     struct inode *inode, *parent;
 978     unsigned int filter = 0;
 979     struct stat st, st_new;
 980     char link[35], *name;
 981     int wd, unix_fd;
 982
 983     if (!inotify_fd)
 984         return 0;
 985
 986     unix_fd = get_unix_fd( dir->fd );
 987
 988     /* check if it's in the list of inodes we want to watch */
 989     if (-1 == fstat( unix_fd, &st_new ))
 990         return 0;
 991     inode = find_inode( st_new.st_dev, st_new.st_ino );
 992     if (inode)
 993         return 0;
 994
 995     /* lookup the parent */
 996     sprintf( link, "/proc/self/fd/%u/..", unix_fd );
 997     if (-1 == stat( link, &st ))
 998         return 0;
 999
1000     /*
1001      * If there's no parent, stop.  We could keep going adding
1002      *  ../ to the path until we hit the root of the tree or
1003      *  find a recursively watched ancestor.
1004      * Assume it's too expensive to search up the tree for now.
1005      */
1006     parent = find_inode( st.st_dev, st.st_ino );
1007     if (!parent)
1008         return 0;
1009
1010     if (parent->wd == -1)
1011         return 0;
1012
1013     filter = filter_from_inode( parent, 1 );
1014     if (!filter)
1015         return 0;
1016
1017     sprintf( link, "/proc/self/fd/%u", unix_fd );
1018     name = get_basename( link );
1019     if (!name)
1020         return 0;
1021     inode = inode_add( parent, st_new.st_dev, st_new.st_ino, name );
1022     free( name );
1023     if (!inode)
1024         return 0;
1025
1026     /* Couldn't find this inode at the start of the function, must be new */
1027     assert( inode->wd == -1 );
1028
1029     wd = inotify_add_dir( link, filter );
1030     if (wd != -1)
1031         inode_set_wd( inode, wd );
1032
1033     return 1;
1034 }
1035
1036 #else
1037
1038 static int init_inotify( void )
1039 {
1040     return 0;
1041 }
1042
1043 static int inotify_adjust_changes( struct dir *dir )
1044 {
1045     return 0;
1046 }
1047
1048 static void free_inode( struct inode *inode )
1049 {
1050     assert( 0 );
1051 }
1052
1053 static int dir_add_to_existing_notify( struct dir *dir )
1054 {
1055     return 0;
1056 }
1057
1058 #endif  /* USE_INOTIFY */
1059
1060 struct object *create_dir_obj( struct fd *fd )
1061 {
1062     struct dir *dir;
1063
1064     dir = alloc_object( &dir_ops );
1065     if (!dir)
1066         return NULL;
1067
1068     list_init( &dir->change_q );
1069     list_init( &dir->change_records );
1070     dir->event = NULL;
1071     dir->filter = 0;
1072     dir->notified = 0;
1073     dir->signaled = 0;
1074     dir->want_data = 0;
1075     dir->inode = NULL;
1076     grab_object( fd );
1077     dir->fd = fd;
1078     set_fd_user( fd, &dir_fd_ops, &dir->obj );
1079
1080     dir_add_to_existing_notify( dir );
1081
1082     return &dir->obj;
1083 }
1084
1085 /* enable change notifications for a directory */
1086 DECL_HANDLER(read_directory_changes)
1087 {
1088     struct event *event = NULL;
1089     struct dir *dir;
1090
1091     if (!req->filter)
1092     {
1093         set_error(STATUS_INVALID_PARAMETER);
1094         return;
1095     }
1096
1097     dir = get_dir_obj( current->process, req->handle, 0 );
1098     if (!dir)
1099         return;
1100
1101     /* possibly send changes through an event flag */
1102     if (req->event)
1103     {
1104         event = get_event_obj( current->process, req->event, EVENT_MODIFY_STATE );
1105         if (!event)
1106             goto end;
1107     }
1108
1109     /* discard the current data, and move onto the next event */
1110     if (dir->event) release_object( dir->event );
1111     dir->event = event;
1112
1113     /* requests don't timeout */
1114     if ( req->io_apc && !create_async( current, NULL, &dir->change_q,
1115                         req->io_apc, req->io_user, req->io_sb ))
1116         return;
1117
1118     /* assign it once */
1119     if (!dir->filter)
1120     {
1121         init_inotify();
1122         insert_change( dir );
1123         dir->filter = req->filter;
1124         dir->subtree = req->subtree;
1125         dir->want_data = req->want_data;
1126     }
1127
1128     /* remove any notifications */
1129     if (dir->signaled>0)
1130         dir->signaled--;
1131
1132     /* clear the event */
1133     if (event)
1134         reset_event( event );
1135
1136     /* setup the real notification */
1137     if (!inotify_adjust_changes( dir ))
1138         dnotify_adjust_changes( dir );
1139
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;
1149     struct dir *dir;
1150
1151     dir = get_dir_obj( current->process, req->handle, 0 );
1152     if (!dir)
1153         return;
1154
1155     if ((record = get_first_change_record( dir )) != NULL)
1156     {
1157         reply->action = record->action;
1158         set_reply_data( record->name, record->len );
1159         free( record );
1160     }
1161     else
1162         set_error( STATUS_NO_DATA_DETECTED );
1163
1164     /* now signal it */
1165     dir->signaled++;
1166     dir_signal_changed( dir );
1167
1168     release_object( dir );
1169 }