1---------------------------------------------------------------------- 21. INTRODUCTION 3 4Modern filesystems feature checksumming of data and metadata to 5protect against data corruption. However, the detection of the 6corruption is done at read time which could potentially be months 7after the data was written. At that point the original data that the 8application tried to write is most likely lost. 9 10The solution is to ensure that the disk is actually storing what the 11application meant it to. Recent additions to both the SCSI family 12protocols (SBC Data Integrity Field, SCC protection proposal) as well 13as SATA/T13 (External Path Protection) try to remedy this by adding 14support for appending integrity metadata to an I/O. The integrity 15metadata (or protection information in SCSI terminology) includes a 16checksum for each sector as well as an incrementing counter that 17ensures the individual sectors are written in the right order. And 18for some protection schemes also that the I/O is written to the right 19place on disk. 20 21Current storage controllers and devices implement various protective 22measures, for instance checksumming and scrubbing. But these 23technologies are working in their own isolated domains or at best 24between adjacent nodes in the I/O path. The interesting thing about 25DIF and the other integrity extensions is that the protection format 26is well defined and every node in the I/O path can verify the 27integrity of the I/O and reject it if corruption is detected. This 28allows not only corruption prevention but also isolation of the point 29of failure. 30 31---------------------------------------------------------------------- 322. THE DATA INTEGRITY EXTENSIONS 33 34As written, the protocol extensions only protect the path between 35controller and storage device. However, many controllers actually 36allow the operating system to interact with the integrity metadata 37(IMD). We have been working with several FC/SAS HBA vendors to enable 38the protection information to be transferred to and from their 39controllers. 40 41The SCSI Data Integrity Field works by appending 8 bytes of protection 42information to each sector. The data + integrity metadata is stored 43in 520 byte sectors on disk. Data + IMD are interleaved when 44transferred between the controller and target. The T13 proposal is 45similar. 46 47Because it is highly inconvenient for operating systems to deal with 48520 (and 4104) byte sectors, we approached several HBA vendors and 49encouraged them to allow separation of the data and integrity metadata 50scatter-gather lists. 51 52The controller will interleave the buffers on write and split them on 53read. This means that Linux can DMA the data buffers to and from 54host memory without changes to the page cache. 55 56Also, the 16-bit CRC checksum mandated by both the SCSI and SATA specs 57is somewhat heavy to compute in software. Benchmarks found that 58calculating this checksum had a significant impact on system 59performance for a number of workloads. Some controllers allow a 60lighter-weight checksum to be used when interfacing with the operating 61system. Emulex, for instance, supports the TCP/IP checksum instead. 62The IP checksum received from the OS is converted to the 16-bit CRC 63when writing and vice versa. This allows the integrity metadata to be 64generated by Linux or the application at very low cost (comparable to 65software RAID5). 66 67The IP checksum is weaker than the CRC in terms of detecting bit 68errors. However, the strength is really in the separation of the data 69buffers and the integrity metadata. These two distinct buffers must 70match up for an I/O to complete. 71 72The separation of the data and integrity metadata buffers as well as 73the choice in checksums is referred to as the Data Integrity 74Extensions. As these extensions are outside the scope of the protocol 75bodies (T10, T13), Oracle and its partners are trying to standardize 76them within the Storage Networking Industry Association. 77 78---------------------------------------------------------------------- 793. KERNEL CHANGES 80 81The data integrity framework in Linux enables protection information 82to be pinned to I/Os and sent to/received from controllers that 83support it. 84 85The advantage to the integrity extensions in SCSI and SATA is that 86they enable us to protect the entire path from application to storage 87device. However, at the same time this is also the biggest 88disadvantage. It means that the protection information must be in a 89format that can be understood by the disk. 90 91Generally Linux/POSIX applications are agnostic to the intricacies of 92the storage devices they are accessing. The virtual filesystem switch 93and the block layer make things like hardware sector size and 94transport protocols completely transparent to the application. 95 96However, this level of detail is required when preparing the 97protection information to send to a disk. Consequently, the very 98concept of an end-to-end protection scheme is a layering violation. 99It is completely unreasonable for an application to be aware whether 100it is accessing a SCSI or SATA disk. 101 102The data integrity support implemented in Linux attempts to hide this 103from the application. As far as the application (and to some extent 104the kernel) is concerned, the integrity metadata is opaque information 105that's attached to the I/O. 106 107The current implementation allows the block layer to automatically 108generate the protection information for any I/O. Eventually the 109intent is to move the integrity metadata calculation to userspace for 110user data. Metadata and other I/O that originates within the kernel 111will still use the automatic generation interface. 112 113Some storage devices allow each hardware sector to be tagged with a 11416-bit value. The owner of this tag space is the owner of the block 115device. I.e. the filesystem in most cases. The filesystem can use 116this extra space to tag sectors as they see fit. Because the tag 117space is limited, the block interface allows tagging bigger chunks by 118way of interleaving. This way, 8*16 bits of information can be 119attached to a typical 4KB filesystem block. 120 121This also means that applications such as fsck and mkfs will need 122access to manipulate the tags from user space. A passthrough 123interface for this is being worked on. 124 125 126---------------------------------------------------------------------- 1274. BLOCK LAYER IMPLEMENTATION DETAILS 128 1294.1 BIO 130 131The data integrity patches add a new field to struct bio when 132CONFIG_BLK_DEV_INTEGRITY is enabled. bio_integrity(bio) returns a 133pointer to a struct bip which contains the bio integrity payload. 134Essentially a bip is a trimmed down struct bio which holds a bio_vec 135containing the integrity metadata and the required housekeeping 136information (bvec pool, vector count, etc.) 137 138A kernel subsystem can enable data integrity protection on a bio by 139calling bio_integrity_alloc(bio). This will allocate and attach the 140bip to the bio. 141 142Individual pages containing integrity metadata can subsequently be 143attached using bio_integrity_add_page(). 144 145bio_free() will automatically free the bip. 146 147 1484.2 BLOCK DEVICE 149 150Because the format of the protection data is tied to the physical 151disk, each block device has been extended with a block integrity 152profile (struct blk_integrity). This optional profile is registered 153with the block layer using blk_integrity_register(). 154 155The profile contains callback functions for generating and verifying 156the protection data, as well as getting and setting application tags. 157The profile also contains a few constants to aid in completing, 158merging and splitting the integrity metadata. 159 160Layered block devices will need to pick a profile that's appropriate 161for all subdevices. blk_integrity_compare() can help with that. DM 162and MD linear, RAID0 and RAID1 are currently supported. RAID4/5/6 163will require extra work due to the application tag. 164 165 166---------------------------------------------------------------------- 1675.0 BLOCK LAYER INTEGRITY API 168 1695.1 NORMAL FILESYSTEM 170 171 The normal filesystem is unaware that the underlying block device 172 is capable of sending/receiving integrity metadata. The IMD will 173 be automatically generated by the block layer at submit_bio() time 174 in case of a WRITE. A READ request will cause the I/O integrity 175 to be verified upon completion. 176 177 IMD generation and verification can be toggled using the 178 179 /sys/block/<bdev>/integrity/write_generate 180 181 and 182 183 /sys/block/<bdev>/integrity/read_verify 184 185 flags. 186 187 1885.2 INTEGRITY-AWARE FILESYSTEM 189 190 A filesystem that is integrity-aware can prepare I/Os with IMD 191 attached. It can also use the application tag space if this is 192 supported by the block device. 193 194 195 int bio_integrity_prep(bio); 196 197 To generate IMD for WRITE and to set up buffers for READ, the 198 filesystem must call bio_integrity_prep(bio). 199 200 Prior to calling this function, the bio data direction and start 201 sector must be set, and the bio should have all data pages 202 added. It is up to the caller to ensure that the bio does not 203 change while I/O is in progress. 204 205 bio_integrity_prep() should only be called if 206 bio_integrity_enabled() returned 1. 207 208 2095.3 PASSING EXISTING INTEGRITY METADATA 210 211 Filesystems that either generate their own integrity metadata or 212 are capable of transferring IMD from user space can use the 213 following calls: 214 215 216 struct bip * bio_integrity_alloc(bio, gfp_mask, nr_pages); 217 218 Allocates the bio integrity payload and hangs it off of the bio. 219 nr_pages indicate how many pages of protection data need to be 220 stored in the integrity bio_vec list (similar to bio_alloc()). 221 222 The integrity payload will be freed at bio_free() time. 223 224 225 int bio_integrity_add_page(bio, page, len, offset); 226 227 Attaches a page containing integrity metadata to an existing 228 bio. The bio must have an existing bip, 229 i.e. bio_integrity_alloc() must have been called. For a WRITE, 230 the integrity metadata in the pages must be in a format 231 understood by the target device with the notable exception that 232 the sector numbers will be remapped as the request traverses the 233 I/O stack. This implies that the pages added using this call 234 will be modified during I/O! The first reference tag in the 235 integrity metadata must have a value of bip->bip_sector. 236 237 Pages can be added using bio_integrity_add_page() as long as 238 there is room in the bip bio_vec array (nr_pages). 239 240 Upon completion of a READ operation, the attached pages will 241 contain the integrity metadata received from the storage device. 242 It is up to the receiver to process them and verify data 243 integrity upon completion. 244 245 2465.4 REGISTERING A BLOCK DEVICE AS CAPABLE OF EXCHANGING INTEGRITY 247 METADATA 248 249 To enable integrity exchange on a block device the gendisk must be 250 registered as capable: 251 252 int blk_integrity_register(gendisk, blk_integrity); 253 254 The blk_integrity struct is a template and should contain the 255 following: 256 257 static struct blk_integrity my_profile = { 258 .name = "STANDARDSBODY-TYPE-VARIANT-CSUM", 259 .generate_fn = my_generate_fn, 260 .verify_fn = my_verify_fn, 261 .tuple_size = sizeof(struct my_tuple_size), 262 .tag_size = <tag bytes per hw sector>, 263 }; 264 265 'name' is a text string which will be visible in sysfs. This is 266 part of the userland API so chose it carefully and never change 267 it. The format is standards body-type-variant. 268 E.g. T10-DIF-TYPE1-IP or T13-EPP-0-CRC. 269 270 'generate_fn' generates appropriate integrity metadata (for WRITE). 271 272 'verify_fn' verifies that the data buffer matches the integrity 273 metadata. 274 275 'tuple_size' must be set to match the size of the integrity 276 metadata per sector. I.e. 8 for DIF and EPP. 277 278 'tag_size' must be set to identify how many bytes of tag space 279 are available per hardware sector. For DIF this is either 2 or 280 0 depending on the value of the Control Mode Page ATO bit. 281 282---------------------------------------------------------------------- 2832007-12-24 Martin K. Petersen <martin.petersen@oracle.com> 284