vserver 2.0 rc7
[linux-2.6.git] / arch / um / drivers / ubd_kern.c
index 9b4820e..2a7f689 100644 (file)
 #include "irq_user.h"
 #include "irq_kern.h"
 #include "ubd_user.h"
-#include "2_5compat.h"
 #include "os.h"
 #include "mem.h"
 #include "mem_kern.h"
 #include "cow.h"
 
-enum ubd_req { UBD_READ, UBD_WRITE, UBD_MMAP };
+enum ubd_req { UBD_READ, UBD_WRITE };
 
 struct io_thread_req {
        enum ubd_req op;
@@ -68,8 +67,6 @@ struct io_thread_req {
        unsigned long sector_mask;
        unsigned long long cow_offset;
        unsigned long bitmap_words[2];
-       int map_fd;
-       unsigned long long map_offset;
        int error;
 };
 
@@ -122,10 +119,6 @@ static int ubd_ioctl(struct inode * inode, struct file * file,
 
 #define MAX_DEV (8)
 
-/* Changed in early boot */
-static int ubd_do_mmap = 0;
-#define UBD_MMAP_BLOCK_SIZE PAGE_SIZE
-
 static struct block_device_operations ubd_blops = {
         .owner         = THIS_MODULE,
         .open          = ubd_open,
@@ -156,6 +149,7 @@ static struct gendisk *fake_gendisk[MAX_DEV];
 static struct openflags global_openflags = OPEN_FLAGS;
 
 struct cow {
+       /* This is the backing file, actually */
        char *file;
        int fd;
        unsigned long *bitmap;
@@ -174,12 +168,6 @@ struct ubd {
        int no_cow;
        struct cow cow;
        struct platform_device pdev;
-
-       int map_writes;
-       int map_reads;
-       int nomap_writes;
-       int nomap_reads;
-       int write_maps;
 };
 
 #define DEFAULT_COW { \
@@ -199,11 +187,6 @@ struct ubd {
        .openflags =            OPEN_FLAGS, \
         .no_cow =               0, \
         .cow =                 DEFAULT_COW, \
-       .map_writes             = 0, \
-       .map_reads              = 0, \
-       .nomap_writes           = 0, \
-       .nomap_reads            = 0, \
-       .write_maps             = 0, \
 }
 
 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
@@ -313,13 +296,6 @@ static int ubd_setup_common(char *str, int *index_out)
                int major;
 
                str++;
-               if(!strcmp(str, "mmap")){
-                       CHOOSE_MODE(printk("mmap not supported by the ubd "
-                                          "driver in tt mode\n"),
-                                   ubd_do_mmap = 1);
-                       return(0);
-               }
-
                if(!strcmp(str, "sync")){
                        global_openflags = of_sync(global_openflags);
                        return(0);
@@ -453,6 +429,22 @@ __uml_help(ubd_setup,
 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
 );
 
+static int udb_setup(char *str)
+{
+       printk("udb%s specified on command line is almost certainly a ubd -> "
+              "udb TYPO\n", str);
+       return(1);
+}
+
+__setup("udb", udb_setup);
+__uml_help(udb_setup,
+"udb\n"
+"    This option is here solely to catch ubd -> udb typos, which can be\n"
+"    to impossible to catch visually unless you specifically look for\n"
+"    them.  The only result of any option starting with 'udb' is an error\n"
+"    in the boot output.\n\n"
+);
+
 static int fakehd_set = 0;
 static int fakehd(char *str)
 {
@@ -507,7 +499,7 @@ static void ubd_handler(void)
 {
        struct io_thread_req req;
        struct request *rq = elv_next_request(ubd_queue);
-       int n, err;
+       int n;
 
        do_ubd = NULL;
        intr_count++;
@@ -521,19 +513,6 @@ static void ubd_handler(void)
                return;
        }
         
-       if((req.op != UBD_MMAP) &&
-          ((req.offset != ((__u64) (rq->sector)) << 9) ||
-           (req.length != (rq->current_nr_sectors) << 9)))
-               panic("I/O op mismatch");
-       
-       if(req.map_fd != -1){
-               err = physmem_subst_mapping(req.buffer, req.map_fd,
-                                           req.map_offset, 1);
-               if(err)
-                       printk("ubd_handler - physmem_subst_mapping failed, "
-                              "err = %d\n", -err);
-       }
-
        ubd_finish(rq, req.error);
        reactivate_fd(thread_fd, UBD_IRQ);      
        do_ubd_request(ubd_queue);
@@ -566,14 +545,10 @@ static int ubd_file_size(struct ubd *dev, __u64 *size_out)
 
 static void ubd_close(struct ubd *dev)
 {
-       if(ubd_do_mmap)
-               physmem_forget_descriptor(dev->fd);
        os_close_file(dev->fd);
        if(dev->cow.file == NULL)
                return;
 
-       if(ubd_do_mmap)
-               physmem_forget_descriptor(dev->cow.fd);
        os_close_file(dev->cow.fd);
        vfree(dev->cow.bitmap);
        dev->cow.bitmap = NULL;
@@ -605,7 +580,11 @@ static int ubd_open_dev(struct ubd *dev)
                }
        }
 
-       if(dev->fd < 0) return(dev->fd);
+       if(dev->fd < 0){
+               printk("Failed to open '%s', errno = %d\n", dev->file,
+                      -dev->fd);
+               return(dev->fd);
+       }
 
        if(dev->cow.file != NULL){
                err = -ENOMEM;
@@ -746,15 +725,9 @@ static int ubd_config(char *str)
 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 {
        struct ubd *dev;
-       char *end;
        int n, len = 0;
 
-       n = simple_strtoul(name, &end, 0);
-       if((*end != '\0') || (end == name)){
-               *error_out = "ubd_get_config : didn't parse device number";
-               return(-1);
-       }
-
+       n = parse_unit(&name);
        if((n >= MAX_DEV) || (n < 0)){
                *error_out = "ubd_get_config : device number out of range";
                return(-1);
@@ -913,10 +886,14 @@ static int ubd_open(struct inode *inode, struct file *filp)
                }
        }
        dev->count++;
-       if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
+       set_disk_ro(disk, !dev->openflags.w);
+
+       /* This should no more be needed. And it didn't work anyway to exclude
+        * read-write remounting of filesystems.*/
+       /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
                if(--dev->count == 0) ubd_close(dev);
                err = -EROFS;
-       }
+       }*/
  out:
        return(err);
 }
@@ -991,97 +968,17 @@ static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
                           req->bitmap_words, bitmap_len);
 }
 
-static int mmap_fd(struct request *req, struct ubd *dev, __u64 offset)
-{
-       __u64 sector;
-       unsigned char *bitmap;
-       int bit, i;
-
-       /* mmap must have been requested on the command line */
-       if(!ubd_do_mmap)
-               return(-1);
-
-       /* The buffer must be page aligned */
-       if(((unsigned long) req->buffer % UBD_MMAP_BLOCK_SIZE) != 0)
-               return(-1);
-
-       /* The request must be a page long */
-       if((req->current_nr_sectors << 9) != PAGE_SIZE)
-               return(-1);
-
-       if(dev->cow.file == NULL)
-               return(dev->fd);
-
-       sector = offset >> 9;
-       bitmap = (unsigned char *) dev->cow.bitmap;
-       bit = ubd_test_bit(sector, bitmap);
-
-       for(i = 1; i < req->current_nr_sectors; i++){
-               if(ubd_test_bit(sector + i, bitmap) != bit)
-                       return(-1);
-       }
-
-       if(bit || (rq_data_dir(req) == WRITE))
-               offset += dev->cow.data_offset;
-
-       /* The data on disk must be page aligned */
-       if((offset % UBD_MMAP_BLOCK_SIZE) != 0)
-               return(-1);
-
-       return(bit ? dev->fd : dev->cow.fd);
-}
-
-static int prepare_mmap_request(struct ubd *dev, int fd, __u64 offset,
-                               struct request *req,
-                               struct io_thread_req *io_req)
-{
-       int err;
-
-       if(rq_data_dir(req) == WRITE){
-               /* Writes are almost no-ops since the new data is already in the
-                * host page cache
-                */
-               dev->map_writes++;
-               if(dev->cow.file != NULL)
-                       cowify_bitmap(io_req->offset, io_req->length,
-                                     &io_req->sector_mask, &io_req->cow_offset,
-                                     dev->cow.bitmap, dev->cow.bitmap_offset,
-                                     io_req->bitmap_words,
-                                     dev->cow.bitmap_len);
-       }
-       else {
-               int w;
-
-               if((dev->cow.file != NULL) && (fd == dev->cow.fd))
-                       w = 0;
-               else w = dev->openflags.w;
-
-               if((dev->cow.file != NULL) && (fd == dev->fd))
-                       offset += dev->cow.data_offset;
-
-               err = physmem_subst_mapping(req->buffer, fd, offset, w);
-               if(err){
-                       printk("physmem_subst_mapping failed, err = %d\n",
-                              -err);
-                       return(1);
-               }
-               dev->map_reads++;
-       }
-       io_req->op = UBD_MMAP;
-       io_req->buffer = req->buffer;
-       return(0);
-}
-
 /* Called with ubd_io_lock held */
 static int prepare_request(struct request *req, struct io_thread_req *io_req)
 {
        struct gendisk *disk = req->rq_disk;
        struct ubd *dev = disk->private_data;
        __u64 offset;
-       int len, fd;
+       int len;
 
        if(req->rq_status == RQ_INACTIVE) return(1);
 
+       /* This should be impossible now */
        if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
                printk("Write attempted on readonly ubd device %s\n", 
                       disk->disk_name);
@@ -1094,34 +991,12 @@ static int prepare_request(struct request *req, struct io_thread_req *io_req)
 
        io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
        io_req->fds[1] = dev->fd;
-       io_req->map_fd = -1;
        io_req->cow_offset = -1;
        io_req->offset = offset;
        io_req->length = len;
        io_req->error = 0;
        io_req->sector_mask = 0;
 
-       fd = mmap_fd(req, dev, io_req->offset);
-       if(fd > 0){
-               /* If mmapping is otherwise OK, but the first access to the
-                * page is a write, then it's not mapped in yet.  So we have
-                * to write the data to disk first, then we can map the disk
-                * page in and continue normally from there.
-                */
-               if((rq_data_dir(req) == WRITE) && !is_remapped(req->buffer)){
-                       io_req->map_fd = dev->fd;
-                       io_req->map_offset = io_req->offset +
-                               dev->cow.data_offset;
-                       dev->write_maps++;
-               }
-               else return(prepare_mmap_request(dev, fd, io_req->offset, req,
-                                                io_req));
-       }
-
-       if(rq_data_dir(req) == READ)
-               dev->nomap_reads++;
-       else dev->nomap_writes++;
-
        io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
        io_req->offsets[0] = 0;
        io_req->offsets[1] = dev->cow.data_offset;
@@ -1209,142 +1084,6 @@ static int ubd_ioctl(struct inode * inode, struct file * file,
        return(-EINVAL);
 }
 
-static int ubd_check_remapped(int fd, unsigned long address, int is_write,
-                             __u64 offset)
-{
-       __u64 bitmap_offset;
-       unsigned long new_bitmap[2];
-       int i, err, n;
-
-       /* If it's not a write access, we can't do anything about it */
-       if(!is_write)
-               return(0);
-
-       /* We have a write */
-       for(i = 0; i < sizeof(ubd_dev) / sizeof(ubd_dev[0]); i++){
-               struct ubd *dev = &ubd_dev[i];
-
-               if((dev->fd != fd) && (dev->cow.fd != fd))
-                       continue;
-
-               /* It's a write to a ubd device */
-
-               if(!dev->openflags.w){
-                       /* It's a write access on a read-only device - probably
-                        * shouldn't happen.  If the kernel is trying to change
-                        * something with no intention of writing it back out,
-                        * then this message will clue us in that this needs
-                        * fixing
-                        */
-                       printk("Write access to mapped page from readonly ubd "
-                              "device %d\n", i);
-                       return(0);
-               }
-
-               /* It's a write to a writeable ubd device - it must be COWed
-                * because, otherwise, the page would have been mapped in
-                * writeable
-                */
-
-               if(!dev->cow.file)
-                       panic("Write fault on writeable non-COW ubd device %d",
-                             i);
-
-               /* It should also be an access to the backing file since the
-                * COW pages should be mapped in read-write
-                */
-
-               if(fd == dev->fd)
-                       panic("Write fault on a backing page of ubd "
-                             "device %d\n", i);
-
-               /* So, we do the write, copying the backing data to the COW
-                * file...
-                */
-
-               err = os_seek_file(dev->fd, offset + dev->cow.data_offset);
-               if(err < 0)
-                       panic("Couldn't seek to %lld in COW file of ubd "
-                             "device %d, err = %d",
-                             offset + dev->cow.data_offset, i, -err);
-
-               n = os_write_file(dev->fd, (void *) address, PAGE_SIZE);
-               if(n != PAGE_SIZE)
-                       panic("Couldn't copy data to COW file of ubd "
-                             "device %d, err = %d", i, -n);
-
-               /* ... updating the COW bitmap... */
-
-               cowify_bitmap(offset, PAGE_SIZE, NULL, &bitmap_offset,
-                             dev->cow.bitmap, dev->cow.bitmap_offset,
-                             new_bitmap, dev->cow.bitmap_len);
-
-               err = os_seek_file(dev->fd, bitmap_offset);
-               if(err < 0)
-                       panic("Couldn't seek to %lld in COW file of ubd "
-                             "device %d, err = %d", bitmap_offset, i, -err);
-
-               n = os_write_file(dev->fd, new_bitmap, sizeof(new_bitmap));
-               if(n != sizeof(new_bitmap))
-                       panic("Couldn't update bitmap  of ubd device %d, "
-                             "err = %d", i, -n);
-
-               /* Maybe we can map the COW page in, and maybe we can't.  If
-                * it is a pre-V3 COW file, we can't, since the alignment will
-                * be wrong.  If it is a V3 or later COW file which has been
-                * moved to a system with a larger page size, then maybe we
-                * can't, depending on the exact location of the page.
-                */
-
-               offset += dev->cow.data_offset;
-
-               /* Remove the remapping, putting the original anonymous page
-                * back.  If the COW file can be mapped in, that is done.
-                * Otherwise, the COW page is read in.
-                */
-
-               if(!physmem_remove_mapping((void *) address))
-                       panic("Address 0x%lx not remapped by ubd device %d",
-                             address, i);
-               if((offset % UBD_MMAP_BLOCK_SIZE) == 0)
-                       physmem_subst_mapping((void *) address, dev->fd,
-                                             offset, 1);
-               else {
-                       err = os_seek_file(dev->fd, offset);
-                       if(err < 0)
-                               panic("Couldn't seek to %lld in COW file of "
-                                     "ubd device %d, err = %d", offset, i,
-                                     -err);
-
-                       n = os_read_file(dev->fd, (void *) address, PAGE_SIZE);
-                       if(n != PAGE_SIZE)
-                               panic("Failed to read page from offset %llx of "
-                                     "COW file of ubd device %d, err = %d",
-                                     offset, i, -n);
-               }
-
-               return(1);
-       }
-
-       /* It's not a write on a ubd device */
-       return(0);
-}
-
-static struct remapper ubd_remapper = {
-       .list   = LIST_HEAD_INIT(ubd_remapper.list),
-       .proc   = ubd_check_remapped,
-};
-
-static int ubd_remapper_setup(void)
-{
-       if(ubd_do_mmap)
-               register_remapper(&ubd_remapper);
-
-       return(0);
-}
-
-__initcall(ubd_remapper_setup);
-
 static int same_backing_files(char *from_cmdline, char *from_cow, char *cow)
 {
        struct uml_stat buf1, buf2;
@@ -1547,15 +1286,6 @@ void do_io(struct io_thread_req *req)
        int err;
        __u64 off;
 
-       if(req->op == UBD_MMAP){
-               /* Touch the page to force the host to do any necessary IO to
-                * get it into memory
-                */
-               n = *((volatile int *) req->buffer);
-               req->error = update_bitmap(req);
-               return;
-       }
-
        nsectors = req->length / req->sectorsize;
        start = 0;
        do {
@@ -1591,8 +1321,7 @@ void do_io(struct io_thread_req *req)
                                }
                        } while((n < len) && (n != 0));
                        if (n < len) memset(&buf[n], 0, len - n);
-               }
-               else {
+               } else {
                        n = os_write_file(req->fds[bit], buf, len);
                        if(n != len){
                                printk("do_io - write failed err = %d "