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 static inline int inotify_init( void )
  98 {
  99     int ret;
 100     __asm__ __volatile__( "int $0x80"
 101                           : "=a" (ret)
 102                           : "0" (SYS_inotify_init));
 103     if (ret<0) { errno = -ret; ret = -1; }
 104     return ret;
 105 }
 106
 107 static inline int inotify_add_watch( int fd, const char *name, unsigned int mask )
 108 {
 109     int ret;
 110     __asm__ __volatile__( "pushl %%ebx;\n\t"
 111                           "movl %2,%%ebx;\n\t"
 112                           "int $0x80;\n\t"
 113                           "popl %%ebx"
 114                           : "=a" (ret) : "0" (SYS_inotify_add_watch),
 115                             "r" (fd), "c" (name), "d" (mask) );
 116     if (ret<0) { errno = -ret; ret = -1; }
 117     return ret;
 118 }
 119
 120 static inline int inotify_rm_watch( int fd, int wd )
 121 {
 122     int ret;
 123     __asm__ __volatile__( "pushl %%ebx;\n\t"
 124                           "movl %2,%%ebx;\n\t"
 125                           "int $0x80;\n\t"
 126                           "popl %%ebx"
 127                           : "=a" (ret) : "0" (SYS_inotify_rm_watch),
 128                             "r" (fd), "c" (wd) );
 129     if (ret<0) { errno = -ret; ret = -1; }
 130     return ret;
 131 }
 132
 133 #define USE_INOTIFY
 134
 135 #endif
 136
 137 struct inode;
 138
 139 static void free_inode( struct inode *inode );
 140
 141 static struct fd *inotify_fd;
 142
 143 struct change_record {
 144     struct list entry;
 145     int action;
 146     int len;
 147     char name[1];
 148 };
 149
 150 struct dir
 151 {
 152     struct object  obj;      /* object header */
 153     struct fd     *fd;       /* file descriptor to the directory */
 154     mode_t         mode;     /* file stat.st_mode */
 155     uid_t          uid;      /* file stat.st_uid */
 156     struct list    entry;    /* entry in global change notifications list */
 157     unsigned int   filter;   /* notification filter */
 158     int            notified; /* SIGIO counter */
 159     int            want_data; /* return change data */
 160     int            subtree;  /* do we want to watch subdirectories? */
 161     struct list    change_records;   /* data for the change */
 162     struct list    in_entry; /* entry in the inode dirs list */
 163     struct inode  *inode;    /* inode of the associated directory */
 164 };
 165
 166 static struct fd *dir_get_fd( struct object *obj );
 167 static struct security_descriptor *dir_get_sd( struct object *obj );
 168 static int dir_set_sd( struct object *obj, const struct security_descriptor *sd,
 169                        unsigned int set_info );
 170 static void dir_dump( struct object *obj, int verbose );
 171 static void dir_destroy( struct object *obj );
 172
 173 static const struct object_ops dir_ops =
 174 {
 175     sizeof(struct dir),       /* size */
 176     dir_dump,                 /* dump */
 177     no_get_type,              /* get_type */
 178     add_queue,                /* add_queue */
 179     remove_queue,             /* remove_queue */
 180     default_fd_signaled,      /* signaled */
 181     no_satisfied,             /* satisfied */
 182     no_signal,                /* signal */
 183     dir_get_fd,               /* get_fd */
 184     default_fd_map_access,    /* map_access */
 185     dir_get_sd,               /* get_sd */
 186     dir_set_sd,               /* set_sd */
 187     no_lookup_name,           /* lookup_name */
 188     no_open_file,             /* open_file */
 189     fd_close_handle,          /* close_handle */
 190     dir_destroy               /* destroy */
 191 };
 192
 193 static int dir_get_poll_events( struct fd *fd );
 194 static enum server_fd_type dir_get_fd_type( struct fd *fd );
 195
 196 static const struct fd_ops dir_fd_ops =
 197 {
 198     dir_get_poll_events,         /* get_poll_events */
 199     default_poll_event,          /* poll_event */
 200     no_flush,                    /* flush */
 201     dir_get_fd_type,             /* get_fd_type */
 202     default_fd_removable,        /* removable */
 203     default_fd_ioctl,            /* ioctl */
 204     default_fd_queue_async,      /* queue_async */
 205     default_fd_async_event,      /* async_event */
 206     default_fd_async_terminated, /* async_terminated */
 207     default_fd_cancel_async      /* cancel_async */
 208 };
 209
 210 static struct list change_list = LIST_INIT(change_list);
 211
 212 static void dnotify_adjust_changes( struct dir *dir )
 213 {
 214 #if defined(F_SETSIG) && defined(F_NOTIFY)
 215     int fd = get_unix_fd( dir->fd );
 216     unsigned int filter = dir->filter;
 217     unsigned int val;
 218     if ( 0 > fcntl( fd, F_SETSIG, SIGIO) )
 219         return;
 220
 221     val = DN_MULTISHOT;
 222     if (filter & FILE_NOTIFY_CHANGE_FILE_NAME)
 223         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 224     if (filter & FILE_NOTIFY_CHANGE_DIR_NAME)
 225         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 226     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 227         val |= DN_ATTRIB;
 228     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 229         val |= DN_MODIFY;
 230     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 231         val |= DN_MODIFY;
 232     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 233         val |= DN_ACCESS;
 234     if (filter & FILE_NOTIFY_CHANGE_CREATION)
 235         val |= DN_CREATE;
 236     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 237         val |= DN_ATTRIB;
 238     fcntl( fd, F_NOTIFY, val );
 239 #endif
 240 }
 241
 242 /* insert change in the global list */
 243 static inline void insert_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_add_head( &change_list, &dir->entry );
 251     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 252 }
 253
 254 /* remove change from the global list */
 255 static inline void remove_change( struct dir *dir )
 256 {
 257     sigset_t sigset;
 258
 259     sigemptyset( &sigset );
 260     sigaddset( &sigset, SIGIO );
 261     sigprocmask( SIG_BLOCK, &sigset, NULL );
 262     list_remove( &dir->entry );
 263     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 264 }
 265
 266 static void dir_dump( struct object *obj, int verbose )
 267 {
 268     struct dir *dir = (struct dir *)obj;
 269     assert( obj->ops == &dir_ops );
 270     fprintf( stderr, "Dirfile fd=%p filter=%08x\n", dir->fd, dir->filter );
 271 }
 272
 273 /* enter here directly from SIGIO signal handler */
 274 void do_change_notify( int unix_fd )
 275 {
 276     struct dir *dir;
 277
 278     /* FIXME: this is O(n) ... probably can be improved */
 279     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 280     {
 281         if (get_unix_fd( dir->fd ) != unix_fd) continue;
 282         interlocked_xchg_add( &dir->notified, 1 );
 283         break;
 284     }
 285 }
 286
 287 /* SIGIO callback, called synchronously with the poll loop */
 288 void sigio_callback(void)
 289 {
 290     struct dir *dir;
 291
 292     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 293     {
 294         if (interlocked_xchg( &dir->notified, 0 ))
 295             fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_NOTIFY_ENUM_DIR );
 296     }
 297 }
 298
 299 static struct fd *dir_get_fd( struct object *obj )
 300 {
 301     struct dir *dir = (struct dir *)obj;
 302     assert( obj->ops == &dir_ops );
 303     return (struct fd *)grab_object( dir->fd );
 304 }
 305
 306 static int get_dir_unix_fd( struct dir *dir )
 307 {
 308     return get_unix_fd( dir->fd );
 309 }
 310
 311 static struct security_descriptor *dir_get_sd( struct object *obj )
 312 {
 313     struct dir *dir = (struct dir *)obj;
 314     int unix_fd;
 315     struct stat st;
 316     struct security_descriptor *sd;
 317     assert( obj->ops == &dir_ops );
 318
 319     unix_fd = get_dir_unix_fd( dir );
 320
 321     if (unix_fd == -1 || fstat( unix_fd, &st ) == -1)
 322         return obj->sd;
 323
 324     /* mode and uid the same? if so, no need to re-generate security descriptor */
 325     if (obj->sd &&
 326         (st.st_mode & (S_IRWXU|S_IRWXO)) == (dir->mode & (S_IRWXU|S_IRWXO)) &&
 327         (st.st_uid == dir->uid))
 328         return obj->sd;
 329
 330     sd = mode_to_sd( st.st_mode,
 331                      security_unix_uid_to_sid( st.st_uid ),
 332                      token_get_primary_group( current->process->token ));
 333     if (!sd) return obj->sd;
 334
 335     dir->mode = st.st_mode;
 336     dir->uid = st.st_uid;
 337     free( obj->sd );
 338     obj->sd = sd;
 339     return sd;
 340 }
 341
 342 static int dir_set_sd( struct object *obj, const struct security_descriptor *sd,
 343                        unsigned int set_info )
 344 {
 345     struct dir *dir = (struct dir *)obj;
 346     const SID *owner;
 347     struct stat st;
 348     mode_t mode;
 349     int unix_fd;
 350
 351     assert( obj->ops == &dir_ops );
 352
 353     unix_fd = get_dir_unix_fd( dir );
 354
 355     if (unix_fd == -1 || fstat( unix_fd, &st ) == -1) return 1;
 356
 357     if (set_info & OWNER_SECURITY_INFORMATION)
 358     {
 359         owner = sd_get_owner( sd );
 360         if (!owner)
 361         {
 362             set_error( STATUS_INVALID_SECURITY_DESCR );
 363             return 0;
 364         }
 365         if (!obj->sd || !security_equal_sid( owner, sd_get_owner( obj->sd ) ))
 366         {
 367             /* FIXME: get Unix uid and call fchown */
 368         }
 369     }
 370     else if (obj->sd)
 371         owner = sd_get_owner( obj->sd );
 372     else
 373         owner = token_get_user( current->process->token );
 374
 375     if (set_info & DACL_SECURITY_INFORMATION)
 376     {
 377         /* keep the bits that we don't map to access rights in the ACL */
 378         mode = st.st_mode & (S_ISUID|S_ISGID|S_ISVTX);
 379         mode |= sd_to_mode( sd, owner );
 380
 381         if (((st.st_mode ^ mode) & (S_IRWXU|S_IRWXG|S_IRWXO)) && fchmod( unix_fd, mode ) == -1)
 382         {
 383             file_set_error();
 384             return 0;
 385         }
 386     }
 387     return 1;
 388 }
 389
 390 static struct change_record *get_first_change_record( struct dir *dir )
 391 {
 392     struct list *ptr = list_head( &dir->change_records );
 393     if (!ptr) return NULL;
 394     list_remove( ptr );
 395     return LIST_ENTRY( ptr, struct change_record, entry );
 396 }
 397
 398 static void dir_destroy( struct object *obj )
 399 {
 400     struct change_record *record;
 401     struct dir *dir = (struct dir *)obj;
 402     assert (obj->ops == &dir_ops);
 403
 404     if (dir->filter)
 405         remove_change( dir );
 406
 407     if (dir->inode)
 408     {
 409         list_remove( &dir->in_entry );
 410         free_inode( dir->inode );
 411     }
 412
 413     while ((record = get_first_change_record( dir ))) free( record );
 414
 415     release_object( dir->fd );
 416
 417     if (inotify_fd && list_empty( &change_list ))
 418     {
 419         release_object( inotify_fd );
 420         inotify_fd = NULL;
 421     }
 422 }
 423
 424 struct dir *get_dir_obj( struct process *process, obj_handle_t handle, unsigned int access )
 425 {
 426     return (struct dir *)get_handle_obj( process, handle, access, &dir_ops );
 427 }
 428
 429 static int dir_get_poll_events( struct fd *fd )
 430 {
 431     return 0;
 432 }
 433
 434 static enum server_fd_type dir_get_fd_type( struct fd *fd )
 435 {
 436     return FD_TYPE_DIR;
 437 }
 438
 439 #ifdef USE_INOTIFY
 440
 441 #define HASH_SIZE 31
 442
 443 struct inode {
 444     struct list ch_entry;    /* entry in the children list */
 445     struct list children;    /* children of this inode */
 446     struct inode *parent;    /* parent of this inode */
 447     struct list dirs;        /* directory handles watching this inode */
 448     struct list ino_entry;   /* entry in the inode hash */
 449     struct list wd_entry;    /* entry in the watch descriptor hash */
 450     dev_t dev;               /* device number */
 451     ino_t ino;               /* device's inode number */
 452     int wd;                  /* inotify's watch descriptor */
 453     char *name;              /* basename name of the inode */
 454 };
 455
 456 static struct list inode_hash[ HASH_SIZE ];
 457 static struct list wd_hash[ HASH_SIZE ];
 458
 459 static int inotify_add_dir( char *path, unsigned int filter );
 460
 461 static struct inode *inode_from_wd( int wd )
 462 {
 463     struct list *bucket = &wd_hash[ wd % HASH_SIZE ];
 464     struct inode *inode;
 465
 466     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, wd_entry )
 467         if (inode->wd == wd)
 468             return inode;
 469
 470     return NULL;
 471 }
 472
 473 static inline struct list *get_hash_list( dev_t dev, ino_t ino )
 474 {
 475     return &inode_hash[ (ino ^ dev) % HASH_SIZE ];
 476 }
 477
 478 static struct inode *find_inode( dev_t dev, ino_t ino )
 479 {
 480     struct list *bucket = get_hash_list( dev, ino );
 481     struct inode *inode;
 482
 483     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, ino_entry )
 484         if (inode->ino == ino && inode->dev == dev)
 485              return inode;
 486
 487     return NULL;
 488 }
 489
 490 static struct inode *create_inode( dev_t dev, ino_t ino )
 491 {
 492     struct inode *inode;
 493
 494     inode = malloc( sizeof *inode );
 495     if (inode)
 496     {
 497         list_init( &inode->children );
 498         list_init( &inode->dirs );
 499         inode->ino = ino;
 500         inode->dev = dev;
 501         inode->wd = -1;
 502         inode->parent = NULL;
 503         inode->name = NULL;
 504         list_add_tail( get_hash_list( dev, ino ), &inode->ino_entry );
 505     }
 506     return inode;
 507 }
 508
 509 static struct inode *get_inode( dev_t dev, ino_t ino )
 510 {
 511     struct inode *inode;
 512
 513     inode = find_inode( dev, ino );
 514     if (inode)
 515         return inode;
 516     return create_inode( dev, ino );
 517 }
 518
 519 static void inode_set_wd( struct inode *inode, int wd )
 520 {
 521     if (inode->wd != -1)
 522         list_remove( &inode->wd_entry );
 523     inode->wd = wd;
 524     list_add_tail( &wd_hash[ wd % HASH_SIZE ], &inode->wd_entry );
 525 }
 526
 527 static void inode_set_name( struct inode *inode, const char *name )
 528 {
 529     free (inode->name);
 530     inode->name = name ? strdup( name ) : NULL;
 531 }
 532
 533 static void free_inode( struct inode *inode )
 534 {
 535     int subtree = 0, watches = 0;
 536     struct inode *tmp, *next;
 537     struct dir *dir;
 538
 539     LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 540     {
 541         subtree |= dir->subtree;
 542         watches++;
 543     }
 544
 545     if (!subtree && !inode->parent)
 546     {
 547         LIST_FOR_EACH_ENTRY_SAFE( tmp, next, &inode->children,
 548                                   struct inode, ch_entry )
 549         {
 550             assert( tmp != inode );
 551             assert( tmp->parent == inode );
 552             free_inode( tmp );
 553         }
 554     }
 555
 556     if (watches)
 557         return;
 558
 559     if (inode->parent)
 560         list_remove( &inode->ch_entry );
 561
 562     /* disconnect remaining children from the parent */
 563     LIST_FOR_EACH_ENTRY_SAFE( tmp, next, &inode->children, struct inode, ch_entry )
 564     {
 565         list_remove( &tmp->ch_entry );
 566         tmp->parent = NULL;
 567     }
 568
 569     if (inode->wd != -1)
 570     {
 571         inotify_rm_watch( get_unix_fd( inotify_fd ), inode->wd );
 572         list_remove( &inode->wd_entry );
 573     }
 574     list_remove( &inode->ino_entry );
 575
 576     free( inode->name );
 577     free( inode );
 578 }
 579
 580 static struct inode *inode_add( struct inode *parent,
 581                                 dev_t dev, ino_t ino, const char *name )
 582 {
 583     struct inode *inode;
 584
 585     inode = get_inode( dev, ino );
 586     if (!inode)
 587         return NULL;
 588
 589     if (!inode->parent)
 590     {
 591         list_add_tail( &parent->children, &inode->ch_entry );
 592         inode->parent = parent;
 593         assert( inode != parent );
 594     }
 595     inode_set_name( inode, name );
 596
 597     return inode;
 598 }
 599
 600 static struct inode *inode_from_name( struct inode *inode, const char *name )
 601 {
 602     struct inode *i;
 603
 604     LIST_FOR_EACH_ENTRY( i, &inode->children, struct inode, ch_entry )
 605         if (i->name && !strcmp( i->name, name ))
 606             return i;
 607     return NULL;
 608 }
 609
 610 static int inotify_get_poll_events( struct fd *fd );
 611 static void inotify_poll_event( struct fd *fd, int event );
 612
 613 static const struct fd_ops inotify_fd_ops =
 614 {
 615     inotify_get_poll_events,     /* get_poll_events */
 616     inotify_poll_event,          /* poll_event */
 617     NULL,                        /* flush */
 618     NULL,                        /* get_fd_type */
 619     NULL,                        /* removable */
 620     NULL,                        /* ioctl */
 621     NULL,                        /* queue_async */
 622     NULL,                        /* async_event */
 623     NULL,                        /* async_terminated */
 624     NULL                         /* cancel_async */
 625 };
 626
 627 static int inotify_get_poll_events( struct fd *fd )
 628 {
 629     return POLLIN;
 630 }
 631
 632 static void inotify_do_change_notify( struct dir *dir, unsigned int action,
 633                                       const char *relpath )
 634 {
 635     struct change_record *record;
 636
 637     assert( dir->obj.ops == &dir_ops );
 638
 639     if (dir->want_data)
 640     {
 641         size_t len = strlen(relpath);
 642         record = malloc( offsetof(struct change_record, name[len]) );
 643         if (!record)
 644             return;
 645
 646         record->action = action;
 647         memcpy( record->name, relpath, len );
 648         record->len = len;
 649
 650         list_add_tail( &dir->change_records, &record->entry );
 651     }
 652
 653     fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_ALERTED );
 654 }
 655
 656 static unsigned int filter_from_event( struct inotify_event *ie )
 657 {
 658     unsigned int filter = 0;
 659
 660     if (ie->mask & (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE))
 661         filter |= FILE_NOTIFY_CHANGE_FILE_NAME | FILE_NOTIFY_CHANGE_DIR_NAME;
 662     if (ie->mask & IN_MODIFY)
 663         filter |= FILE_NOTIFY_CHANGE_SIZE | FILE_NOTIFY_CHANGE_LAST_WRITE;
 664     if (ie->mask & IN_ATTRIB)
 665         filter |= FILE_NOTIFY_CHANGE_ATTRIBUTES | FILE_NOTIFY_CHANGE_SECURITY;
 666     if (ie->mask & IN_ACCESS)
 667         filter |= FILE_NOTIFY_CHANGE_LAST_ACCESS;
 668     if (ie->mask & IN_CREATE)
 669         filter |= FILE_NOTIFY_CHANGE_CREATION;
 670
 671     return filter;
 672 }
 673
 674 /* scan up the parent directories for watches */
 675 static unsigned int filter_from_inode( struct inode *inode, int is_parent )
 676 {
 677     unsigned int filter = 0;
 678     struct dir *dir;
 679
 680     /* combine filters from parents watching subtrees */
 681     while (inode)
 682     {
 683         LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 684             if (dir->subtree || !is_parent)
 685                 filter |= dir->filter;
 686         is_parent = 1;
 687         inode = inode->parent;
 688     }
 689
 690     return filter;
 691 }
 692
 693 static char *inode_get_path( struct inode *inode, int sz )
 694 {
 695     struct list *head;
 696     char *path;
 697     int len;
 698
 699     if (!inode)
 700         return NULL;
 701
 702     head = list_head( &inode->dirs );
 703     if (head)
 704     {
 705         int unix_fd = get_unix_fd( LIST_ENTRY( head, struct dir, in_entry )->fd );
 706         path = malloc ( 32 + sz );
 707         if (path)
 708             sprintf( path, "/proc/self/fd/%u/", unix_fd );
 709         return path;
 710     }
 711
 712     if (!inode->name)
 713         return NULL;
 714
 715     len = strlen( inode->name );
 716     path = inode_get_path( inode->parent, sz + len + 1 );
 717     if (!path)
 718         return NULL;
 719
 720     strcat( path, inode->name );
 721     strcat( path, "/" );
 722
 723     return path;
 724 }
 725
 726 static int inode_check_dir( struct inode *parent, const char *name )
 727 {
 728     char *path;
 729     unsigned int filter;
 730     struct inode *inode;
 731     struct stat st;
 732     int wd = -1, r = -1;
 733
 734     path = inode_get_path( parent, strlen(name) );
 735     if (!path)
 736         return r;
 737
 738     strcat( path, name );
 739
 740     r = stat( path, &st );
 741     if (r < 0) goto end;
 742
 743     if (!S_ISDIR(st.st_mode))
 744     {
 745         r = 0;
 746         goto end;
 747     }
 748
 749     r = 1;
 750
 751     filter = filter_from_inode( parent, 1 );
 752     if (!filter)
 753         goto end;
 754
 755     inode = inode_add( parent, st.st_dev, st.st_ino, name );
 756     if (!inode || inode->wd != -1)
 757         goto end;
 758
 759     wd = inotify_add_dir( path, filter );
 760     if (wd != -1)
 761         inode_set_wd( inode, wd );
 762     else
 763         free_inode( inode );
 764
 765 end:
 766     free( path );
 767     return r;
 768 }
 769
 770 static int prepend( char **path, const char *segment )
 771 {
 772     int extra;
 773     char *p;
 774
 775     extra = strlen( segment ) + 1;
 776     if (*path)
 777     {
 778         int len = strlen( *path ) + 1;
 779         p = realloc( *path, len + extra );
 780         if (!p) return 0;
 781         memmove( &p[ extra ], p, len );
 782         p[ extra - 1 ] = '/';
 783         memcpy( p, segment, extra - 1 );
 784     }
 785     else
 786     {
 787         p = malloc( extra );
 788         if (!p) return 0;
 789         memcpy( p, segment, extra );
 790     }
 791
 792     *path = p;
 793
 794     return 1;
 795 }
 796
 797 static void inotify_notify_all( struct inotify_event *ie )
 798 {
 799     unsigned int filter, action;
 800     struct inode *inode, *i;
 801     char *path = NULL;
 802     struct dir *dir;
 803
 804     inode = inode_from_wd( ie->wd );
 805     if (!inode)
 806     {
 807         fprintf( stderr, "no inode matches %d\n", ie->wd);
 808         return;
 809     }
 810
 811     filter = filter_from_event( ie );
 812
 813     if (ie->mask & IN_CREATE)
 814     {
 815         switch (inode_check_dir( inode, ie->name ))
 816         {
 817         case 1:
 818             filter &= ~FILE_NOTIFY_CHANGE_FILE_NAME;
 819             break;
 820         case 0:
 821             filter &= ~FILE_NOTIFY_CHANGE_DIR_NAME;
 822             break;
 823         default:
 824             break;
 825             /* Maybe the file disappeared before we could check it? */
 826         }
 827         action = FILE_ACTION_ADDED;
 828     }
 829     else if (ie->mask & IN_DELETE)
 830         action = FILE_ACTION_REMOVED;
 831     else
 832         action = FILE_ACTION_MODIFIED;
 833
 834     /*
 835      * Work our way up the inode hierarchy
 836      *  extending the relative path as we go
 837      *  and notifying all recursive watches.
 838      */
 839     if (!prepend( &path, ie->name ))
 840         return;
 841
 842     for (i = inode; i; i = i->parent)
 843     {
 844         LIST_FOR_EACH_ENTRY( dir, &i->dirs, struct dir, in_entry )
 845             if ((filter & dir->filter) && (i==inode || dir->subtree))
 846                 inotify_do_change_notify( dir, action, path );
 847
 848         if (!i->name || !prepend( &path, i->name ))
 849             break;
 850     }
 851
 852     free( path );
 853
 854     if (ie->mask & IN_DELETE)
 855     {
 856         i = inode_from_name( inode, ie->name );
 857         if (i)
 858             free_inode( i );
 859     }
 860 }
 861
 862 static void inotify_poll_event( struct fd *fd, int event )
 863 {
 864     int r, ofs, unix_fd;
 865     char buffer[0x1000];
 866     struct inotify_event *ie;
 867
 868     unix_fd = get_unix_fd( fd );
 869     r = read( unix_fd, buffer, sizeof buffer );
 870     if (r < 0)
 871     {
 872         fprintf(stderr,"inotify_poll_event(): inotify read failed!\n");
 873         return;
 874     }
 875
 876     for( ofs = 0; ofs < r - offsetof(struct inotify_event, name); )
 877     {
 878         ie = (struct inotify_event*) &buffer[ofs];
 879         if (!ie->len)
 880             break;
 881         ofs += offsetof( struct inotify_event, name[ie->len] );
 882         if (ofs > r) break;
 883         inotify_notify_all( ie );
 884     }
 885 }
 886
 887 static inline struct fd *create_inotify_fd( void )
 888 {
 889     int unix_fd;
 890
 891     unix_fd = inotify_init();
 892     if (unix_fd<0)
 893         return NULL;
 894     return create_anonymous_fd( &inotify_fd_ops, unix_fd, NULL, 0 );
 895 }
 896
 897 static int map_flags( unsigned int filter )
 898 {
 899     unsigned int mask;
 900
 901     /* always watch these so we can track subdirectories in recursive watches */
 902     mask = (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE | IN_DELETE_SELF);
 903
 904     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 905         mask |= IN_ATTRIB;
 906     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 907         mask |= IN_MODIFY;
 908     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 909         mask |= IN_MODIFY;
 910     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 911         mask |= IN_ACCESS;
 912     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 913         mask |= IN_ATTRIB;
 914
 915     return mask;
 916 }
 917
 918 static int inotify_add_dir( char *path, unsigned int filter )
 919 {
 920     int wd = inotify_add_watch( get_unix_fd( inotify_fd ),
 921                                 path, map_flags( filter ) );
 922     if (wd != -1)
 923         set_fd_events( inotify_fd, POLLIN );
 924     return wd;
 925 }
 926
 927 static int init_inotify( void )
 928 {
 929     int i;
 930
 931     if (inotify_fd)
 932         return 1;
 933
 934     inotify_fd = create_inotify_fd();
 935     if (!inotify_fd)
 936         return 0;
 937
 938     for (i=0; i<HASH_SIZE; i++)
 939     {
 940         list_init( &inode_hash[i] );
 941         list_init( &wd_hash[i] );
 942     }
 943
 944     return 1;
 945 }
 946
 947 static int inotify_adjust_changes( struct dir *dir )
 948 {
 949     unsigned int filter;
 950     struct inode *inode;
 951     struct stat st;
 952     char path[32];
 953     int wd, unix_fd;
 954
 955     if (!inotify_fd)
 956         return 0;
 957
 958     unix_fd = get_unix_fd( dir->fd );
 959
 960     inode = dir->inode;
 961     if (!inode)
 962     {
 963         /* check if this fd is already being watched */
 964         if (-1 == fstat( unix_fd, &st ))
 965             return 0;
 966
 967         inode = get_inode( st.st_dev, st.st_ino );
 968         if (!inode)
 969             inode = create_inode( st.st_dev, st.st_ino );
 970         if (!inode)
 971             return 0;
 972         list_add_tail( &inode->dirs, &dir->in_entry );
 973         dir->inode = inode;
 974     }
 975
 976     filter = filter_from_inode( inode, 0 );
 977
 978     sprintf( path, "/proc/self/fd/%u", unix_fd );
 979     wd = inotify_add_dir( path, filter );
 980     if (wd == -1) return 0;
 981
 982     inode_set_wd( inode, wd );
 983
 984     return 1;
 985 }
 986
 987 static char *get_basename( const char *link )
 988 {
 989     char *buffer, *name = NULL;
 990     int r, n = 0x100;
 991
 992     while (1)
 993     {
 994         buffer = malloc( n );
 995         if (!buffer) return NULL;
 996
 997         r = readlink( link, buffer, n );
 998         if (r < 0)
 999             break;
1000
1001         if (r < n)
1002         {
1003             name = buffer;
1004             break;
1005         }
1006         free( buffer );
1007         n *= 2;
1008     }
1009
1010     if (name)
1011     {
1012         while (r > 0 && name[ r - 1 ] == '/' )
1013             r--;
1014         name[ r ] = 0;
1015
1016         name = strrchr( name, '/' );
1017         if (name)
1018             name = strdup( &name[1] );
1019     }
1020
1021     free( buffer );
1022     return name;
1023 }
1024
1025 static int dir_add_to_existing_notify( struct dir *dir )
1026 {
1027     struct inode *inode, *parent;
1028     unsigned int filter = 0;
1029     struct stat st, st_new;
1030     char link[35], *name;
1031     int wd, unix_fd;
1032
1033     if (!inotify_fd)
1034         return 0;
1035
1036     unix_fd = get_unix_fd( dir->fd );
1037
1038     /* check if it's in the list of inodes we want to watch */
1039     if (-1 == fstat( unix_fd, &st_new ))
1040         return 0;
1041     inode = find_inode( st_new.st_dev, st_new.st_ino );
1042     if (inode)
1043         return 0;
1044
1045     /* lookup the parent */
1046     sprintf( link, "/proc/self/fd/%u/..", unix_fd );
1047     if (-1 == stat( link, &st ))
1048         return 0;
1049
1050     /*
1051      * If there's no parent, stop.  We could keep going adding
1052      *  ../ to the path until we hit the root of the tree or
1053      *  find a recursively watched ancestor.
1054      * Assume it's too expensive to search up the tree for now.
1055      */
1056     parent = find_inode( st.st_dev, st.st_ino );
1057     if (!parent)
1058         return 0;
1059
1060     if (parent->wd == -1)
1061         return 0;
1062
1063     filter = filter_from_inode( parent, 1 );
1064     if (!filter)
1065         return 0;
1066
1067     sprintf( link, "/proc/self/fd/%u", unix_fd );
1068     name = get_basename( link );
1069     if (!name)
1070         return 0;
1071     inode = inode_add( parent, st_new.st_dev, st_new.st_ino, name );
1072     free( name );
1073     if (!inode)
1074         return 0;
1075
1076     /* Couldn't find this inode at the start of the function, must be new */
1077     assert( inode->wd == -1 );
1078
1079     wd = inotify_add_dir( link, filter );
1080     if (wd != -1)
1081         inode_set_wd( inode, wd );
1082
1083     return 1;
1084 }
1085
1086 #else
1087
1088 static int init_inotify( void )
1089 {
1090     return 0;
1091 }
1092
1093 static int inotify_adjust_changes( struct dir *dir )
1094 {
1095     return 0;
1096 }
1097
1098 static void free_inode( struct inode *inode )
1099 {
1100     assert( 0 );
1101 }
1102
1103 static int dir_add_to_existing_notify( struct dir *dir )
1104 {
1105     return 0;
1106 }
1107
1108 #endif  /* USE_INOTIFY */
1109
1110 struct object *create_dir_obj( struct fd *fd, unsigned int access, mode_t mode )
1111 {
1112     struct dir *dir;
1113
1114     dir = alloc_object( &dir_ops );
1115     if (!dir)
1116         return NULL;
1117
1118     list_init( &dir->change_records );
1119     dir->filter = 0;
1120     dir->notified = 0;
1121     dir->want_data = 0;
1122     dir->inode = NULL;
1123     grab_object( fd );
1124     dir->fd = fd;
1125     dir->mode = mode;
1126     dir->uid  = ~(uid_t)0;
1127     set_fd_user( fd, &dir_fd_ops, &dir->obj );
1128
1129     dir_add_to_existing_notify( dir );
1130
1131     return &dir->obj;
1132 }
1133
1134 /* enable change notifications for a directory */
1135 DECL_HANDLER(read_directory_changes)
1136 {
1137     struct dir *dir;
1138     struct async *async;
1139
1140     if (!req->filter)
1141     {
1142         set_error(STATUS_INVALID_PARAMETER);
1143         return;
1144     }
1145
1146     dir = get_dir_obj( current->process, req->async.handle, 0 );
1147     if (!dir)
1148         return;
1149
1150     /* requests don't timeout */
1151     if (!(async = fd_queue_async( dir->fd, &req->async, ASYNC_TYPE_WAIT ))) goto end;
1152
1153     /* assign it once */
1154     if (!dir->filter)
1155     {
1156         init_inotify();
1157         insert_change( dir );
1158         dir->filter = req->filter;
1159         dir->subtree = req->subtree;
1160         dir->want_data = req->want_data;
1161     }
1162
1163     /* if there's already a change in the queue, send it */
1164     if (!list_empty( &dir->change_records ))
1165         fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_ALERTED );
1166
1167     /* setup the real notification */
1168     if (!inotify_adjust_changes( dir ))
1169         dnotify_adjust_changes( dir );
1170
1171     release_object( async );
1172     set_error(STATUS_PENDING);
1173
1174 end:
1175     release_object( dir );
1176 }
1177
1178 DECL_HANDLER(read_change)
1179 {
1180     struct change_record *record;
1181     struct dir *dir;
1182
1183     dir = get_dir_obj( current->process, req->handle, 0 );
1184     if (!dir)
1185         return;
1186
1187     if ((record = get_first_change_record( dir )) != NULL)
1188     {
1189         reply->action = record->action;
1190         set_reply_data( record->name, record->len );
1191         free( record );
1192     }
1193     else
1194         set_error( STATUS_NO_DATA_DETECTED );
1195
1196     release_object( dir );
1197 }