linux/generic: sync 2.6.37 patches
[openwrt.git] / target / linux / generic / patches-2.6.37 / 065-rootfs_split.patch
1 --- a/drivers/mtd/Kconfig
2 +++ b/drivers/mtd/Kconfig
3 @@ -53,6 +53,16 @@ config MTD_PARTITIONS
4           devices. Partitioning on NFTL 'devices' is a different - that's the
5           'normal' form of partitioning used on a block device.
6  
7 +config MTD_ROOTFS_ROOT_DEV
8 +       bool "Automatically set 'rootfs' partition to be root filesystem"
9 +       depends on MTD_PARTITIONS
10 +       default y
11 +
12 +config MTD_ROOTFS_SPLIT
13 +       bool "Automatically split 'rootfs' partition for squashfs"
14 +       depends on MTD_PARTITIONS
15 +       default y
16 +
17  config MTD_REDBOOT_PARTS
18         tristate "RedBoot partition table parsing"
19         depends on MTD_PARTITIONS
20 --- a/drivers/mtd/mtdpart.c
21 +++ b/drivers/mtd/mtdpart.c
22 @@ -29,6 +29,8 @@
23  #include <linux/kmod.h>
24  #include <linux/mtd/mtd.h>
25  #include <linux/mtd/partitions.h>
26 +#include <linux/root_dev.h>
27 +#include <linux/magic.h>
28  #include <linux/err.h>
29  
30  /* Our partition linked list */
31 @@ -48,7 +50,7 @@ struct mtd_part {
32   * the pointer to that structure with this macro.
33   */
34  #define PART(x)  ((struct mtd_part *)(x))
35 -
36 +#define IS_PART(mtd) (mtd->read == part_read)
37  
38  /*
39   * MTD methods which simply translate the effective address and pass through
40 @@ -618,6 +620,150 @@ int mtd_del_partition(struct mtd_info *m
41  }
42  EXPORT_SYMBOL_GPL(mtd_del_partition);
43  
44 +#ifdef CONFIG_MTD_ROOTFS_SPLIT
45 +#define ROOTFS_SPLIT_NAME "rootfs_data"
46 +#define ROOTFS_REMOVED_NAME "<removed>"
47 +
48 +struct squashfs_super_block {
49 +       __le32 s_magic;
50 +       __le32 pad0[9];
51 +       __le64 bytes_used;
52 +};
53 +
54 +
55 +static int split_squashfs(struct mtd_info *master, int offset, int *split_offset)
56 +{
57 +       struct squashfs_super_block sb;
58 +       int len, ret;
59 +
60 +       ret = master->read(master, offset, sizeof(sb), &len, (void *) &sb);
61 +       if (ret || (len != sizeof(sb))) {
62 +               printk(KERN_ALERT "split_squashfs: error occured while reading "
63 +                       "from \"%s\"\n", master->name);
64 +               return -EINVAL;
65 +       }
66 +
67 +       if (SQUASHFS_MAGIC != le32_to_cpu(sb.s_magic) ) {
68 +               printk(KERN_ALERT "split_squashfs: no squashfs found in \"%s\"\n",
69 +                       master->name);
70 +               *split_offset = 0;
71 +               return 0;
72 +       }
73 +
74 +       if (le64_to_cpu((sb.bytes_used)) <= 0) {
75 +               printk(KERN_ALERT "split_squashfs: squashfs is empty in \"%s\"\n",
76 +                       master->name);
77 +               *split_offset = 0;
78 +               return 0;
79 +       }
80 +
81 +       len = (u32) le64_to_cpu(sb.bytes_used);
82 +       len += (offset & 0x000fffff);
83 +       len +=  (master->erasesize - 1);
84 +       len &= ~(master->erasesize - 1);
85 +       len -= (offset & 0x000fffff);
86 +       *split_offset = offset + len;
87 +
88 +       return 0;
89 +}
90 +
91 +static int split_rootfs_data(struct mtd_info *master, struct mtd_info *rpart, const struct mtd_partition *part)
92 +{
93 +       struct mtd_partition *dpart;
94 +       struct mtd_part *slave = NULL;
95 +       int split_offset = 0;
96 +       int ret;
97 +
98 +       ret = split_squashfs(master, part->offset, &split_offset);
99 +       if (ret)
100 +               return ret;
101 +
102 +       if (split_offset <= 0)
103 +               return 0;
104 +
105 +       dpart = kmalloc(sizeof(*part)+sizeof(ROOTFS_SPLIT_NAME)+1, GFP_KERNEL);
106 +       if (dpart == NULL) {
107 +               printk(KERN_INFO "split_squashfs: no memory for partition \"%s\"\n",
108 +                       ROOTFS_SPLIT_NAME);
109 +               return -ENOMEM;
110 +       }
111 +
112 +       memcpy(dpart, part, sizeof(*part));
113 +       dpart->name = (unsigned char *)&dpart[1];
114 +       strcpy(dpart->name, ROOTFS_SPLIT_NAME);
115 +
116 +       dpart->size -= split_offset - dpart->offset;
117 +       dpart->offset = split_offset;
118 +
119 +       if (dpart == NULL)
120 +               return 1;
121 +
122 +       printk(KERN_INFO "mtd: partition \"%s\" created automatically, ofs=%llX, len=%llX \n",
123 +               ROOTFS_SPLIT_NAME, dpart->offset, dpart->size);
124 +
125 +       slave = allocate_partition(master, dpart, 0, split_offset);
126 +       if (!slave) {
127 +               kfree(dpart);
128 +               return -ENOMEM;
129 +       }
130 +       rpart->split = &slave->mtd;
131 +
132 +       return 0;
133 +}
134 +
135 +static int refresh_rootfs_split(struct mtd_info *mtd)
136 +{
137 +       struct mtd_partition tpart;
138 +       struct mtd_part *part;
139 +       char *name;
140 +       //int index = 0;
141 +       int offset, size;
142 +       int ret;
143 +
144 +       part = PART(mtd);
145 +
146 +       /* check for the new squashfs offset first */
147 +       ret = split_squashfs(part->master, part->offset, &offset);
148 +       if (ret)
149 +               return ret;
150 +
151 +       if ((offset > 0) && !mtd->split) {
152 +               printk(KERN_INFO "%s: creating new split partition for \"%s\"\n", __func__, mtd->name);
153 +               /* if we don't have a rootfs split partition, create a new one */
154 +               tpart.name = (char *) mtd->name;
155 +               tpart.size = mtd->size;
156 +               tpart.offset = part->offset;
157 +
158 +               return split_rootfs_data(part->master, &part->mtd, &tpart);
159 +       } else if ((offset > 0) && mtd->split) {
160 +               /* update the offsets of the existing partition */
161 +               size = mtd->size + part->offset - offset;
162 +
163 +               part = PART(mtd->split);
164 +               part->offset = offset;
165 +               part->mtd.size = size;
166 +               printk(KERN_INFO "%s: %s partition \"" ROOTFS_SPLIT_NAME "\", offset: 0x%06x (0x%06x)\n",
167 +                       __func__, (!strcmp(part->mtd.name, ROOTFS_SPLIT_NAME) ? "updating" : "creating"),
168 +                       (u32) part->offset, (u32) part->mtd.size);
169 +               name = kmalloc(sizeof(ROOTFS_SPLIT_NAME) + 1, GFP_KERNEL);
170 +               strcpy(name, ROOTFS_SPLIT_NAME);
171 +               part->mtd.name = name;
172 +       } else if ((offset <= 0) && mtd->split) {
173 +               printk(KERN_INFO "%s: removing partition \"%s\"\n", __func__, mtd->split->name);
174 +
175 +               /* mark existing partition as removed */
176 +               part = PART(mtd->split);
177 +               name = kmalloc(sizeof(ROOTFS_SPLIT_NAME) + 1, GFP_KERNEL);
178 +               strcpy(name, ROOTFS_REMOVED_NAME);
179 +               part->mtd.name = name;
180 +               part->offset = 0;
181 +               part->mtd.size = 0;
182 +       }
183 +
184 +       return 0;
185 +}
186 +#endif /* CONFIG_MTD_ROOTFS_SPLIT */
187 +
188  /*
189   * This function, given a master MTD object and a partition table, creates
190   * and registers slave MTD objects which are bound to the master according to
191 @@ -633,7 +779,7 @@ int add_mtd_partitions(struct mtd_info *
192  {
193         struct mtd_part *slave;
194         uint64_t cur_offset = 0;
195 -       int i;
196 +       int i, ret;
197  
198         printk(KERN_NOTICE "Creating %d MTD partitions on \"%s\":\n", nbparts, master->name);
199  
200 @@ -648,6 +794,21 @@ int add_mtd_partitions(struct mtd_info *
201  
202                 add_mtd_device(&slave->mtd);
203  
204 +               if (!strcmp(parts[i].name, "rootfs")) {
205 +#ifdef CONFIG_MTD_ROOTFS_ROOT_DEV
206 +                       if (ROOT_DEV == 0) {
207 +                               printk(KERN_NOTICE "mtd: partition \"rootfs\" "
208 +                                       "set to be root filesystem\n");
209 +                               ROOT_DEV = MKDEV(MTD_BLOCK_MAJOR, slave->mtd.index);
210 +                       }
211 +#endif
212 +#ifdef CONFIG_MTD_ROOTFS_SPLIT
213 +                       ret = split_rootfs_data(master, &slave->mtd, &parts[i]);
214 +                       /* if (ret == 0)
215 +                        *      j++; */
216 +#endif
217 +               }
218 +
219                 cur_offset = slave->offset + slave->mtd.size;
220         }
221  
222 @@ -655,6 +816,32 @@ int add_mtd_partitions(struct mtd_info *
223  }
224  EXPORT_SYMBOL(add_mtd_partitions);
225  
226 +int refresh_mtd_partitions(struct mtd_info *mtd)
227 +{
228 +       int ret = 0;
229 +
230 +       if (IS_PART(mtd)) {
231 +               struct mtd_part *part;
232 +               struct mtd_info *master;
233 +
234 +               part = PART(mtd);
235 +               master = part->master;
236 +               if (master->refresh_device)
237 +                       ret = master->refresh_device(master);
238 +       }
239 +
240 +       if (!ret && mtd->refresh_device)
241 +               ret = mtd->refresh_device(mtd);
242 +
243 +#ifdef CONFIG_MTD_ROOTFS_SPLIT
244 +       if (!ret && IS_PART(mtd) && !strcmp(mtd->name, "rootfs"))
245 +               refresh_rootfs_split(mtd);
246 +#endif
247 +
248 +       return 0;
249 +}
250 +EXPORT_SYMBOL_GPL(refresh_mtd_partitions);
251 +
252  static DEFINE_SPINLOCK(part_parser_lock);
253  static LIST_HEAD(part_parsers);
254  
255 --- a/drivers/mtd/devices/block2mtd.c
256 +++ b/drivers/mtd/devices/block2mtd.c
257 @@ -30,6 +30,8 @@ struct block2mtd_dev {
258         struct block_device *blkdev;
259         struct mtd_info mtd;
260         struct mutex write_mutex;
261 +       rwlock_t bdev_mutex;
262 +       char devname[0];
263  };
264  
265  
266 @@ -82,6 +84,12 @@ static int block2mtd_erase(struct mtd_in
267         size_t len = instr->len;
268         int err;
269  
270 +       read_lock(&dev->bdev_mutex);
271 +       if (!dev->blkdev) {
272 +               err = -EINVAL;
273 +               goto done;
274 +       }
275 +
276         instr->state = MTD_ERASING;
277         mutex_lock(&dev->write_mutex);
278         err = _block2mtd_erase(dev, from, len);
279 @@ -93,6 +101,10 @@ static int block2mtd_erase(struct mtd_in
280                 instr->state = MTD_ERASE_DONE;
281  
282         mtd_erase_callback(instr);
283 +
284 +done:
285 +       read_unlock(&dev->bdev_mutex);
286 +
287         return err;
288  }
289  
290 @@ -104,10 +116,14 @@ static int block2mtd_read(struct mtd_inf
291         struct page *page;
292         int index = from >> PAGE_SHIFT;
293         int offset = from & (PAGE_SIZE-1);
294 -       int cpylen;
295 +       int cpylen, err = 0;
296 +
297 +       read_lock(&dev->bdev_mutex);
298 +       if (!dev->blkdev || (from > mtd->size)) {
299 +               err = -EINVAL;
300 +               goto done;
301 +       }
302  
303 -       if (from > mtd->size)
304 -               return -EINVAL;
305         if (from + len > mtd->size)
306                 len = mtd->size - from;
307  
308 @@ -122,10 +138,14 @@ static int block2mtd_read(struct mtd_inf
309                 len = len - cpylen;
310  
311                 page = page_read(dev->blkdev->bd_inode->i_mapping, index);
312 -               if (!page)
313 -                       return -ENOMEM;
314 -               if (IS_ERR(page))
315 -                       return PTR_ERR(page);
316 +               if (!page) {
317 +                       err = -ENOMEM;
318 +                       goto done;
319 +               }
320 +               if (IS_ERR(page)) {
321 +                       err = PTR_ERR(page);
322 +                       goto done;
323 +               }
324  
325                 memcpy(buf, page_address(page) + offset, cpylen);
326                 page_cache_release(page);
327 @@ -136,7 +156,10 @@ static int block2mtd_read(struct mtd_inf
328                 offset = 0;
329                 index++;
330         }
331 -       return 0;
332 +
333 +done:
334 +       read_unlock(&dev->bdev_mutex);
335 +       return err;
336  }
337  
338  
339 @@ -188,12 +211,22 @@ static int block2mtd_write(struct mtd_in
340                 size_t *retlen, const u_char *buf)
341  {
342         struct block2mtd_dev *dev = mtd->priv;
343 -       int err;
344 +       int err = 0;
345 +
346 +       read_lock(&dev->bdev_mutex);
347 +       if (!dev->blkdev) {
348 +               err = -EINVAL;
349 +               goto done;
350 +       }
351  
352         if (!len)
353 -               return 0;
354 -       if (to >= mtd->size)
355 -               return -ENOSPC;
356 +               goto done;
357 +
358 +       if (to >= mtd->size) {
359 +               err = -ENOSPC;
360 +               goto done;
361 +       }
362 +
363         if (to + len > mtd->size)
364                 len = mtd->size - to;
365  
366 @@ -202,6 +235,9 @@ static int block2mtd_write(struct mtd_in
367         mutex_unlock(&dev->write_mutex);
368         if (err > 0)
369                 err = 0;
370 +
371 +done:
372 +       read_unlock(&dev->bdev_mutex);
373         return err;
374  }
375  
376 @@ -210,52 +246,29 @@ static int block2mtd_write(struct mtd_in
377  static void block2mtd_sync(struct mtd_info *mtd)
378  {
379         struct block2mtd_dev *dev = mtd->priv;
380 -       sync_blockdev(dev->blkdev);
381 -       return;
382 -}
383 -
384 -
385 -static void block2mtd_free_device(struct block2mtd_dev *dev)
386 -{
387 -       if (!dev)
388 -               return;
389 -
390 -       kfree(dev->mtd.name);
391  
392 -       if (dev->blkdev) {
393 -               invalidate_mapping_pages(dev->blkdev->bd_inode->i_mapping,
394 -                                       0, -1);
395 -               close_bdev_exclusive(dev->blkdev, FMODE_READ|FMODE_WRITE);
396 -       }
397 +       read_lock(&dev->bdev_mutex);
398 +       if (dev->blkdev)
399 +               sync_blockdev(dev->blkdev);
400 +       read_unlock(&dev->bdev_mutex);
401  
402 -       kfree(dev);
403 +       return;
404  }
405  
406  
407 -/* FIXME: ensure that mtd->size % erase_size == 0 */
408 -static struct block2mtd_dev *add_device(char *devname, int erase_size, const char *mtdname)
409 +static int _open_bdev(struct block2mtd_dev *dev)
410  {
411         struct block_device *bdev;
412 -       struct block2mtd_dev *dev;
413 -       struct mtd_partition *part;
414 -       char *name;
415 -
416 -       if (!devname)
417 -               return NULL;
418 -
419 -       dev = kzalloc(sizeof(struct block2mtd_dev), GFP_KERNEL);
420 -       if (!dev)
421 -               return NULL;
422  
423         /* Get a handle on the device */
424 -       bdev = open_bdev_exclusive(devname, FMODE_READ|FMODE_WRITE, NULL);
425 +       bdev = open_bdev_exclusive(dev->devname, FMODE_READ|FMODE_WRITE, NULL);
426  #ifndef MODULE
427         if (IS_ERR(bdev)) {
428  
429                 /* We might not have rootfs mounted at this point. Try
430                    to resolve the device name by other means. */
431  
432 -               dev_t devt = name_to_dev_t(devname);
433 +               dev_t devt = name_to_dev_t(dev->devname);
434                 if (devt) {
435                         bdev = open_by_devnum(devt, FMODE_WRITE | FMODE_READ);
436                 }
437 @@ -263,17 +276,98 @@ static struct block2mtd_dev *add_device(
438  #endif
439  
440         if (IS_ERR(bdev)) {
441 -               ERROR("error: cannot open device %s", devname);
442 -               goto devinit_err;
443 +               ERROR("error: cannot open device %s", dev->devname);
444 +               return 1;
445         }
446         dev->blkdev = bdev;
447  
448         if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
449                 ERROR("attempting to use an MTD device as a block device");
450 -               goto devinit_err;
451 +               return 1;
452         }
453  
454 +       return 0;
455 +}
456 +
457 +static void _close_bdev(struct block2mtd_dev *dev)
458 +{
459 +       struct block_device *bdev;
460 +
461 +       if (!dev->blkdev)
462 +               return;
463 +
464 +       bdev = dev->blkdev;
465 +       invalidate_mapping_pages(dev->blkdev->bd_inode->i_mapping, 0, -1);
466 +       close_bdev_exclusive(dev->blkdev, FMODE_READ|FMODE_WRITE);
467 +       dev->blkdev = NULL;
468 +}
469 +
470 +static void block2mtd_free_device(struct block2mtd_dev *dev)
471 +{
472 +       if (!dev)
473 +               return;
474 +
475 +       kfree(dev->mtd.name);
476 +       _close_bdev(dev);
477 +       kfree(dev);
478 +}
479 +
480 +
481 +static int block2mtd_refresh(struct mtd_info *mtd)
482 +{
483 +       struct block2mtd_dev *dev = mtd->priv;
484 +       struct block_device *bdev;
485 +       dev_t devt;
486 +       int err = 0;
487 +
488 +       /* no other mtd function can run at this point */
489 +       write_lock(&dev->bdev_mutex);
490 +
491 +       /* get the device number for the whole disk */
492 +       devt = MKDEV(MAJOR(dev->blkdev->bd_dev), 0);
493 +
494 +       /* close the old block device */
495 +       _close_bdev(dev);
496 +
497 +       /* open the whole disk, issue a partition rescan, then */
498 +       bdev = open_by_devnum(devt, FMODE_WRITE | FMODE_READ);
499 +       if (!bdev || !bdev->bd_disk)
500 +               err = -EINVAL;
501 +#ifndef CONFIG_MTD_BLOCK2MTD_MODULE
502 +       else
503 +               err = rescan_partitions(bdev->bd_disk, bdev);
504 +#endif
505 +       if (bdev)
506 +               close_bdev_exclusive(bdev, FMODE_READ|FMODE_WRITE);
507 +
508 +       /* try to open the partition block device again */
509 +       _open_bdev(dev);
510 +       write_unlock(&dev->bdev_mutex);
511 +
512 +       return err;
513 +}
514 +
515 +/* FIXME: ensure that mtd->size % erase_size == 0 */
516 +static struct block2mtd_dev *add_device(char *devname, int erase_size, char *mtdname)
517 +{
518 +       struct block2mtd_dev *dev;
519 +       struct mtd_partition *part;
520 +       char *name;
521 +
522 +       if (!devname)
523 +               return NULL;
524 +
525 +       dev = kzalloc(sizeof(struct block2mtd_dev) + strlen(devname) + 1, GFP_KERNEL);
526 +       if (!dev)
527 +               return NULL;
528 +
529 +       strcpy(dev->devname, devname);
530 +
531 +       if (_open_bdev(dev))
532 +               goto devinit_err;
533 +
534         mutex_init(&dev->write_mutex);
535 +       rwlock_init(&dev->bdev_mutex);
536  
537         /* Setup the MTD structure */
538         /* make the name contain the block device in */
539 @@ -298,6 +392,7 @@ static struct block2mtd_dev *add_device(
540         dev->mtd.read = block2mtd_read;
541         dev->mtd.priv = dev;
542         dev->mtd.owner = THIS_MODULE;
543 +       dev->mtd.refresh_device = block2mtd_refresh;
544  
545         part = kzalloc(sizeof(struct mtd_partition), GFP_KERNEL);
546         part->name = dev->mtd.name;
547 --- a/drivers/mtd/mtdchar.c
548 +++ b/drivers/mtd/mtdchar.c
549 @@ -841,6 +841,13 @@ static int mtd_ioctl(struct file *file,
550                 file->f_pos = 0;
551                 break;
552         }
553 +#ifdef CONFIG_MTD_PARTITIONS
554 +       case MTDREFRESH:
555 +       {
556 +               ret = refresh_mtd_partitions(mtd);
557 +               break;
558 +       }
559 +#endif
560  
561         case OTPGETREGIONCOUNT:
562         case OTPGETREGIONINFO:
563 --- a/include/linux/mtd/mtd.h
564 +++ b/include/linux/mtd/mtd.h
565 @@ -125,6 +125,7 @@ struct nand_ecclayout {
566         struct nand_oobfree oobfree[MTD_MAX_OOBFREE_ENTRIES_LARGE];
567  };
568  
569 +struct mtd_info;
570  struct mtd_info {
571         u_char type;
572         uint32_t flags;
573 @@ -266,6 +267,9 @@ struct mtd_info {
574         struct device dev;
575         int usecount;
576  
577 +       int (*refresh_device)(struct mtd_info *mtd);
578 +       struct mtd_info *split;
579 +
580         /* If the driver is something smart, like UBI, it may need to maintain
581          * its own reference counting. The below functions are only for driver.
582          * The driver may register its callbacks. These callbacks are not
583 --- a/include/linux/mtd/partitions.h
584 +++ b/include/linux/mtd/partitions.h
585 @@ -34,12 +34,14 @@
586   * erasesize aligned (e.g. use MTDPART_OFS_NEXTBLK).
587   */
588  
589 +struct mtd_partition;
590  struct mtd_partition {
591         char *name;                     /* identifier string */
592         uint64_t size;                  /* partition size */
593         uint64_t offset;                /* offset within the master MTD space */
594         uint32_t mask_flags;            /* master MTD flags to mask out for this partition */
595         struct nand_ecclayout *ecclayout;       /* out of band layout for this partition (NAND only) */
596 +       int (*refresh_partition)(struct mtd_info *);
597  };
598  
599  #define MTDPART_OFS_NXTBLK     (-2)
600 @@ -51,6 +53,7 @@ struct mtd_info;
601  
602  int add_mtd_partitions(struct mtd_info *, const struct mtd_partition *, int);
603  int del_mtd_partitions(struct mtd_info *);
604 +int refresh_mtd_partitions(struct mtd_info *);
605  
606  /*
607   * Functions dealing with the various ways of partitioning the space
608 --- a/include/mtd/mtd-abi.h
609 +++ b/include/mtd/mtd-abi.h
610 @@ -127,6 +127,7 @@ struct otp_info {
611  #define MEMWRITEOOB64          _IOWR('M', 21, struct mtd_oob_buf64)
612  #define MEMREADOOB64           _IOWR('M', 22, struct mtd_oob_buf64)
613  #define MEMISLOCKED            _IOR('M', 23, struct erase_info_user)
614 +#define MTDREFRESH             _IO('M', 23)
615  
616  /*
617   * Obsolete legacy interface. Keep it in order not to break userspace